读书笔记 on Last DBA

Book Notes — Designing Data-Intensive Applications (2nd Edition)

Fri, 20 Sep 2024 00:00:00 +0000

DDIA-v2 Chinese edition: https://github.com/Vonng/ddia/tree/v2

After finishing DDIA-v2, I couldn’t put it down. Everything data-related is explained with such clarity — why is it like this? What’s the current state? What problems does this have? The observations and ideas are incredibly incisive and concise. Even the nautical-chart-style diagrams at the start of each chapter are fascinating.

Note: This article is essentially a collection of excerpts from the original work, with almost none of my own thoughts or ideas. I’ve simply plucked out the parts I love most. Some knowledge I’ve already mastered and some topics too remote are skipped!

Ch1: Trade-offs in Data System Architecture
#

OLTP & OLAP
#

The distinction between OLTP and analytics is not always clear-cut, but the following table lists some typical characteristics:

Attribute	Transactional Systems (OLTP)	Analytical Systems (OLAP)
Primary read pattern	Point queries (fetch individual records by key)	Aggregation over a large number of records
Primary write pattern	Create, update, and delete individual records	Bulk import (ETL) or event stream
Human user example	End users of web/mobile applications	Internal analysts, for decision support
Machine use example	Check whether an action is authorized	Detect fraud/abuse patterns
Query type	Fixed set of queries, predefined by the application	Analysts can issue arbitrary queries
Data representation	Latest state of data (current point in time)	History of events over time
Dataset size	GB, TB	TB, PB

A data warehouse is a separate database where analysts can query freely without affecting OLTP operations. Data warehouses typically store data in a very different way from OLTP databases, optimized for the query types common in analytics. The process of getting data into the data warehouse is called Extract–Transform–Load (ETL). Some database systems offer Hybrid Transaction/Analytical Processing (HTAP), aiming to enable both OLTP and analytics in a single system without ETL from one system to another. Despite the existence of HTAP, the separation between transactional and analytical systems remains common due to their differing goals and requirements. In particular, it is considered good practice for each business system to have its own database, resulting in hundreds of independent operational databases; on the other hand, an enterprise typically has only one data warehouse, allowing business analysts to combine data from several business systems in a single query. A data lake is a centralized data repository that holds any data potentially useful for analysis, sourced from business systems through ETL processes. Unlike a data warehouse, a data lake contains only files and imposes no specific file format or data model. Data warehouses typically use the relational data model and are queried via SQL. A data lakehouse goes beyond a standalone data warehouse by enabling typical data warehouse workloads (SQL queries and business analytics) as well as data science/machine learning workloads to run directly on files in the data lake. This architecture is called a data lakehouse. It requires a query execution engine and a metadata (e.g., schema management) layer to extend the file storage of the data lake. Apache Hive, Spark SQL, Presto, and Trino are examples of this approach.

Cloud Services vs. Self-Hosting
#

The pros and cons of cloud services: Using cloud services, rather than running comparable software yourself, is essentially outsourcing the operation of that software to a cloud provider. There are strong arguments both for and against using cloud services.

Advantages:

When you use the cloud, you still need an operations team, but outsourcing basic system administration can free your team to focus on higher-level problems.
Cloud services are especially valuable if your system load varies significantly over time. If you provision machines to handle peak load but those computing resources sit idle most of the time, your system becomes less cost-effective.
Compared to physical machines, cloud instances can be provisioned faster and come in a wider variety of sizes.

Disadvantages:

The biggest drawback of cloud services is that you have no control over them.
If you already have experience setting up and operating the required systems and your load is fairly predictable (i.e., the number of machines you need won’t fluctuate dramatically), it is typically cheaper to buy your own machines and run the software yourself.
If the service lacks a feature you need, your only option is to politely ask the vendor whether they’ll add it; you usually can’t implement it yourself.
If the service goes down, you can only wait for it to recover.
If you use the service in a way that triggers a bug or causes performance issues, it’s very difficult to diagnose the problem. With software you run yourself, you can obtain performance metrics and debugging information from the business system to understand its behavior, and you can inspect server logs. But with vendor-hosted services, you typically don’t have access to this internal information.
Moreover, if the service shuts down or becomes unacceptably expensive, or if the vendor decides to change its product in a way you don’t like, you’re at their mercy — continuing to run an old version of the software is usually not an option, so you’ll be forced to migrate to another service. This risk can be mitigated if there are alternative services offering compatible APIs, but for many cloud services, there is no standard API, which increases switching costs and makes vendor lock-in a real problem.
Latency-critical applications such as high-frequency trading require complete control over hardware, making the cloud a poor choice for such businesses.

Cloud-Native
#

Category	Self-Hosted Systems	Cloud-Native Systems
Transactional/OLTP	MySQL, PostgreSQL, MongoDB	AWS Aurora, Azure SQL DB Hyperscale, Google Cloud Spanner
Analytical/OLAP	Teradata, ClickHouse, Spark	Snowflake, Google BigQuery, Azure Synapse Analytics

The key idea behind cloud-native services is not only to use computing resources managed by the business system but also to build on top of lower-level cloud services to create higher-level services. For example:

Object storage services like Amazon S3, Azure Blob Storage, and Cloudflare R2 store large files. They provide a more limited API than a typical filesystem (basic file reads and writes), but their advantage is hiding the underlying physical machines: the service automatically distributes data across many machines, so you don’t need to worry about running out of disk space on any single machine. Even if some machines or their disks fail entirely, no data is lost.
Many other services are in turn built on top of object storage and other cloud services: for example, Snowflake is a cloud-based analytical database (data warehouse) that relies on S3 for data storage, and some services are further built on top of Snowflake.

Cloud-native systems are typically multi-tenant, meaning they don’t provision separate machines for each customer. Instead, data and computation from several different customers are handled by the same service on shared hardware. Multi-tenancy enables better hardware utilization, easier scalability, and simpler management for cloud providers.

Operations in the Cloud Era
#

Traditionally, the people managing an organization’s server-side data infrastructure were called database administrators (DBAs) or system administrators (sysadmins). In recent years, many organizations have attempted to integrate software development and operations roles into a single team jointly responsible for backend services and data infrastructure; the DevOps philosophy has guided this trend. Site Reliability Engineers (SREs) represent Google’s implementation of this philosophy.

The DevOps/SRE philosophy emphasizes:

Automation — preferring repeatable processes over one-off manual tasks,
Preferring ephemeral virtual machines and services over long-running servers,
Promoting frequent application updates,
Learning from incidents,
Preserving organizational knowledge about systems even as individual personnel come and go.

The operations team at an infrastructure company focuses on the details of reliably delivering services to a large number of customers, while the customers of the service spend as little time and energy on infrastructure as possible. Beyond the traditional need for capacity planning, adopting cloud services may be easier and faster than running your own infrastructure. While the cloud is changing the role of operations, the need for operations remains urgent.

Ch2: Defining Non-Functional Requirements
#

Hardware and Software Faults
#

In large-scale systems, hardware faults happen frequently enough that they become part of normal system operation:

About 2-5% of disk hard drives fail each year; in a storage cluster with 10,000 disks, we can therefore expect on average one disk failure per day.
About 0.5-1% of solid-state drives (SSDs) fail each year. Uncorrectable errors occur about once per drive per year.
About one in 1,000 machines has a CPU core that occasionally computes incorrect results.
Data in RAM can also be corrupted, due to random events like cosmic rays or permanent physical defects. Additionally, certain pathological memory access patterns can flip bits with high probability.
Other hardware components such as power supplies, RAID controllers, and memory modules also fail.
An entire data center can become unavailable (e.g., due to power outages or network misconfiguration) or even permanently destroyed (e.g., fire or flood).

Software faults are often unpredictable and, because they are correlated across nodes, can cause more system failures than hardware faults:

A bug that causes all application server instances to crash upon receiving a specific bad input. For example, the leap second on June 30, 2012, caused many applications to hang simultaneously due to a bug in the Linux kernel.
A runaway process that exhausts some shared resource — CPU time, memory, disk space, or network bandwidth.
A service that the system depends on becomes slow, unresponsive, or starts returning incorrect responses.
Cascading failures, where a small fault in one component triggers a fault in another, which triggers further faults.

Operational configuration errors are the leading cause of service outages, while hardware faults (server or network) account for only 10-25% of service outages.

Scalability Principles
#

A good general principle for scalability is to decompose the system into small components that can operate relatively independently. This is the basic principle behind microservices. However, the challenge lies in knowing where to draw the line between things that belong together and things that should be separate.

If a single-machine database can do the job, it may be preferable to a complex distributed setup. A system with five services is simpler than one with fifty services. Good architecture often involves a mix of approaches.

Operations
#

An operations team is critical to keeping software systems running smoothly. The typical responsibilities of a good operations team include (and go beyond) the following:

Monitoring system health and quickly restoring service when it degrades.
Tracking down the causes of problems, such as system failures or performance degradation.
Keeping software and platforms up to date, including security patches.
Understanding interactions between systems to avoid damaging changes before they cause harm.
Anticipating future problems and addressing them before they occur (e.g., capacity planning).
Establishing good practices for deployment, configuration, and management, and writing supporting tools.
Performing complex maintenance tasks, such as migrating applications from one platform to another.
Maintaining system security during configuration changes.
Defining workflows to make operations predictable and maintain production environment stability.
Preserving organizational knowledge about systems as personnel come and go.

Good operability means easier day-to-day work, allowing the operations team to focus on high-value tasks. Data systems can make routine tasks easier in various ways:

Providing good monitoring with visibility into the system’s internal state and runtime behavior.
Offering good support for automation, integrating the system with standardized tools.
Avoiding dependence on a single machine (allowing machines to be taken down for maintenance while the overall system continues running uninterrupted).
Providing good documentation and an easy-to-understand operational model (“if you do X, Y will happen”).
Providing good default behavior but also allowing administrators to freely override defaults when needed.
Self-healing when possible, but also allowing administrators to manually control system state when needed.
Predictable behavior, minimizing surprises.

Some aspects of operations can and should be automated, but setting up correctly functioning automation in the first place still depends on humans.

Systems with too strong an individual stamp cannot succeed. When the initial design is complete and relatively stable, the real testing begins as different people test it in their own ways. — Donald Knuth

Ch3: Data Models and Query Languages
#

Most applications are built by layering one data model on top of another.

As an application developer, you observe the real world (with people, organizations, goods, actions, money flows, sensors, etc.) and model it in terms of objects or data structures and APIs that manipulate those data structures. These structures are typically specific to your application.
When you want to store these data structures, you express them in a general-purpose data model, such as JSON or XML documents, tables in a relational database, or vertices and edges in a graph. These data models are the subject of this chapter.
The engineers who build your database software decided on a way to represent that JSON/relational/graph data as bytes in memory, on disk, or on the network. This representation may allow the data to be queried, searched, manipulated, and processed in various ways. We’ll discuss these storage engine designs in a later chapter.
At an even lower level, hardware engineers have figured out how to represent bytes in terms of electric currents, light pulses, magnetic fields, and so on.

SQL & NoSQL
#

Databases can execute declarative queries in parallel across multiple CPU cores and machines, without you needing to worry about how to implement that parallelism. Implementing such parallel execution yourself in hand-coded algorithms would be an enormous undertaking.

The relational model, despite being half a century old, remains an important data model for many applications — especially in data warehousing and business analytics, where relational star or snowflake schemas and SQL queries are ubiquitous. However, in other domains, several alternatives to relational data have become popular:

The document model targets use cases where data comes in the form of self-contained JSON documents and relationships between documents are rare.
The graph data model goes in the opposite direction, targeting use cases where anything can be related to everything, and queries may need to traverse multiple hops to find data of interest (this can be expressed using recursive queries in Cypher, SPARQL, or Datalog).
The dataframe generalizes relational data into a large number of columns, building a bridge between databases and the multidimensional arrays that form the foundation of most machine learning, statistical data analysis, and scientific computing.

Databases also tend to expand into adjacent domains by adding support for other data models: for example, relational databases have added support for document data in the form of JSON columns, document databases have added relational-like joins, and support for graph data in SQL is gradually improving.

Ch4: Storage and Indexing
#

Hash Indexes
#

Key-value stores are quite similar to the dictionary type found in most programming languages, which is typically implemented using a hash map or hash table.

Generally, the hash map of a hash index is kept entirely in memory. Data values can use more space than available memory because the required portion can be loaded from disk with a single disk seek.

Drawbacks of hash indexes:

In principle, a hash map can be maintained on disk. Unfortunately, disk-based hash maps struggle to perform well. They require a large amount of random-access I/O, are expensive to grow when exhausted, and require tedious logic to resolve hash collisions.
Range queries are inefficient. For example, you can’t easily scan all keys between kitty00000 and kitty99999 — you must look up each key individually in the hash map.

B-Tree Indexes
#

B-tree indexes have been around since 1970 and are widely accepted and used in the industry. This section is familiar to most readers — skipped.

SSTables & LSM Trees
#

In hash indexes, the order of key-value pairs doesn’t matter. But we can require that the sequence of key-value pairs be sorted by key. This format is called a Sorted String Table, or SSTable.

Compared to log segments using hash indexes, SSTables have several major advantages:

Even if the file is larger than available memory, merging segments remains simple and efficient. The approach is like the one used in merge sort algorithms: you start reading multiple input files side by side, look at the first key in each file, copy the lowest key (according to the sort order) to the output file, and repeat. This produces a new merged segment file, also sorted by key.

To find a particular key in the file, you no longer need to keep an index of all keys in memory. You still need an in-memory index to tell you the offsets for some of the keys, but it can be sparse: one key per several kilobytes of segment file is sufficient, because several kilobytes can be scanned very quickly.

Using these data structures, you can insert keys in any order and read them back in sorted order. Now we can make our storage engine work as follows:

When a new write comes in, add it to an in-memory balanced tree data structure (e.g., a red-black tree). This in-memory tree is sometimes called a memtable.
When the memtable becomes larger than some threshold (typically a few megabytes), write it out to disk as an SSTable file. This can be done efficiently because the tree already maintains key-value pairs sorted by key. The new SSTable file becomes the most recent segment of the database. While that SSTable is being written to disk, new writes can continue on a new memtable instance.
When a read request comes in, first try to find the key in the memtable, then in the most recent on-disk segment, then in the next older segment, and so on.
From time to time, run a merging and compaction process in the background to combine segment files and discard overwritten or deleted values.

The algorithm described here is essentially the technique used by LevelDB and RocksDB, key-value storage engine libraries designed to be embedded in other applications. Similar storage engines are used in Cassandra and HBase, and all of them were inspired by Google’s Bigtable paper (which introduced the terms SSTable and memtable).

In-Memory Databases
#

In-memory databases: As RAM becomes cheaper, the argument that RAM costs more per GB is eroding. Many datasets are not that large, so keeping them entirely in memory is quite feasible, including potentially distributed across multiple machines. This has led to the development of in-memory databases. Losing data when restarting a computer may be acceptable. Durability can also be achieved through special hardware (e.g., battery-backed RAM), by writing a change log to disk, by periodically writing snapshots to disk, or by replicating the in-memory state to other machines.

The typical in-memory database Redis provides weak durability through asynchronous writes to disk. Other in-memory databases include Memcached, VoltDB, MemSQL, Oracle TimesTen, and RAMCloud.

Counterintuitively, the performance advantage of in-memory databases does not come from avoiding disk reads. Instead, they are faster because they avoid the overhead of encoding in-memory data structures into on-disk data structures.

Materialized Views and OLAP
#

Think of SQL functions like COUNT, SUM, AVG, MIN, or MAX. If the same aggregations are used by many different queries, it may be wasteful to process the raw data each time. Why not cache some of the most frequently used counts or sums? One way to create such a cache is a Materialized View.

When the underlying data changes, a materialized view needs to be updated because it is a denormalized copy of the data. The database can do this automatically, but such updates make writes more expensive, which is why materialized views are not commonly used in OLTP databases. In read-heavy data warehouses, they may make more sense because warehouses don’t have many small, frequent updates.

The advantage of a materialized data cube is that it can make certain queries extremely fast because they have already been effectively precomputed. For example, if you want to know the total sales per store, you just look at the total along the appropriate dimension without scanning millions of rows of raw data.

The disadvantage of a data cube is that it lacks the flexibility of querying raw data. For example, there is no way to compute what proportion of sales came from items costing over $100, because price is not one of the dimensions. Therefore, most data warehouses try to keep as much raw data as possible and use aggregate data (like data cubes) only as a performance boost for certain queries.

Column-Oriented Storage
#

The idea behind column-oriented storage is simple: instead of storing all the values from one row together, store all the values from each column together. Column-oriented storage is easiest to understand in the relational data model, but it applies equally to non-relational data. For example, Parquet is a column-oriented storage format that supports a document data model based on Google’s Dremel.

These optimizations (column compression, sorting, etc.) make sense in data warehouses, where the workload consists mainly of large read-only queries run by analysts. Column-oriented storage, compression, and sorting all help read those queries faster. However, their drawback is that writes become more difficult.

Ch5: Encoding and Evolution
#

REST vs. RPC
#

Servers expose APIs over the network, and clients can connect to servers to make requests to those APIs. The API exposed by a server is called a service. Download data via GET requests, submit data to the server via POST requests.

When a service uses HTTP as the underlying communication protocol, it can be called a web service. There are two popular approaches to web services: REST and SOAP. REST is not a protocol but a design philosophy based on HTTP principles. APIs designed according to REST principles are called RESTful.

Remote Procedure Calls (RPC) are very different from local function calls:

Local function calls are predictable and succeed or fail based only on parameters under your control. Network requests are unpredictable: requests or responses may be lost due to network problems, or the remote machine may be slow or unavailable.
A local function call either returns a result, throws an exception, or never returns (because it enters an infinite loop or the process crashes). A network request has another possible outcome: it may return with no result due to a timeout.
And so on.

REST seems to be the dominant style for public APIs, while RPC frameworks mainly focus on requests between services owned by the same organization, typically within the same data center.

Ch6: Replication
#

Replication logs, failover, single-leader mode — the content is relatively straightforward. Skipped.

Multi-Leader Replication
#

Multi-leader replication is often a retrofitted feature in many databases, so it frequently has subtle configuration pitfalls and often interacts unexpectedly with other database features. For example, auto-increment primary keys, triggers, and integrity constraints can all cause trouble. Therefore, multi-leader replication is often considered dangerous territory and should be avoided whenever possible.

However, multi-leader replication does have certain advantages, such as distributing write I/O, disaster recovery, and reducing network overhead in multi-region deployments (local writes), etc.

Write conflicts: The biggest problem with multi-leader replication is the potential for write conflicts, and resolving them is quite tricky. In principle, conflict detection could be made synchronous — i.e., wait for writes to be replicated to all replicas before telling the user the write succeeded. But this may defeat the purpose of multi-leader: if you want synchronous conflict detection, you might as well just use single-leader replication.

Resolving multi-leader write conflicts:

Avoid conflicts. For example, have the application control that users only edit their own data.
Converge to consistency:
- Last Write Wins (LWW). Write by timestamp — may result in data loss.
- Priority writes. Higher-priority writes win — may result in data loss.
- Extra code. Preserve conflict information and write custom conflict resolution code.

Real-time collaborative editing applications allow multiple people to edit a document simultaneously — Etherpad and Google Docs are mature examples. Databases are still very young in the area of multi-leader writes.

Multi-leader write conflicts in databases are mostly resolved or avoided at the application level. The following are relatively mature areas of write conflict research for reference:

Conflict-free Replicated Data Types (CRDTs) are data structures such as sets, maps, ordered lists, and counters that can be concurrently edited by multiple users and resolve conflicts automatically in a reasonable way. Some CRDTs have been implemented in Riak 2.0.
Mergeable Persistent Data Structures explicitly track history, similar to the Git version control system, and use three-way merge functions (whereas CRDTs use two-way merges).
Operational Transformation (OT) is the conflict resolution algorithm behind collaborative editing applications like Etherpad and Google Docs. It is designed specifically for concurrent editing of ordered lists, such as lists of characters that make up a text document.

Ch7: Partitioning
#

Range Partitioning and Hash Partitioning
#

The drawback of range partitioning is that certain access patterns can lead to hot spots. If the primary key is a timestamp, partitions correspond to time ranges, and all writes will go to the same partition (i.e., today’s partition), which may become overloaded with writes while other partitions sit idle. You can use something other than the timestamp as the first part of the primary key to scatter the hot spot, but the drawback is that range queries won’t benefit.

Hash partitioning can mitigate the risk of skew and hot spots. For the purpose of partitioning, the hash function doesn’t need to be a cryptographically strong algorithm. The drawback of hash partitioning is that by partitioning by key hash, we lose a great property of key-range partitioning: the ability to efficiently execute range queries.

Hash partitioning can help reduce hot spots. But it cannot eliminate them entirely. For example, on a social media site, a celebrity user with millions of followers doing something can trigger a storm. This event can cause a large number of writes to the same key (the key might be the celebrity’s user ID or the ID of the action being commented on). Hash strategies don’t help here, because the hash of two identical IDs is still the same. If a primary key is very hot, a simple workaround is to add a random number at the beginning or end of the primary key. Just a two-digit decimal random number can scatter the primary key into 100 different primary keys, thus stored in different partitions. In any case, it’s about scattering hot spots, and you need to consider side effects such as the impact on range queries.

Ch8: Transactions
#

ACID, BASE
#

ACID is actually a very old definition. Due to the later discovery of many “anomalies,” a system claiming to guarantee ACID can’t actually articulate what exactly it guarantees. Whatever the case, ACID remains deeply ingrained — it represents the most fundamental principles of transactions. Conversely, systems that don’t meet the ACID criteria are sometimes called BASE, which stands for Basically Available, Soft State, and Eventual Consistency. BASE is a concept commonly mentioned in the NoSQL world.

The definition of BASE is even fuzzier than ACID. A simple, easy-to-understand, easy-to-remember theory of BASE: BASE (which means “alkali” in chemistry) is the opposite of ACID (which means “acid”).

You can think of it simply this way:


Relational databases	Non-relational databases
Transactions	No transactions
ACID	BASE
SQL	NoSQL

Atomicity and isolation within ACID are relatively easy to understand. The concept of consistency is actually quite vague and doesn’t seem closely related to the database itself. A quote in the book is very classic:

Joe Hellerstein pointed out that in Härder and Reuter’s paper, “the C in ACID” was “tossed in to make the acronym work,” and at the time, nobody cared much about consistency.

And the definition of isolation is very fuzzy. The industrial practice of serializability has also been stagnant. Transaction isolation can be described as “a mess,” but if serializability is a panacea, why does no one use it? Refer to this article: The History of Transactions and SSI

Anomalies in non-serializable isolation levels generally only manifest under high concurrency; databases with low concurrency rarely encounter problems.
When anomalies do occur, some applications may not notice them or may detect them but find them unimportant.
Data may be anomalous, but the application may simply return an error and enter an anomaly-handling routine.
Cost is too high. Not only is the development cost of serializable isolation levels high for databases, but applications also need adaptation costs for serializability. Just understanding this complex theory is no easy task.
Higher isolation levels lose some performance. Massive rework may be thankless; applications need to choose between “high concurrency” and “no anomalies.”
Businesses develop based on mechanisms, not rules. Businesses have somewhat adapted to the anomalies of weaker isolation levels, especially Read Committed.

Summed up in one sentence: It’s not like it’s unusable!

Pessimistic and Optimistic Transaction Models
#

Two-phase locking is a so-called pessimistic concurrency control mechanism: it is based on the principle that if something might go wrong (e.g., another transaction holding a lock), it’s better to wait until the situation is safe before proceeding. It’s like a mutex used to protect data structures in multi-threaded programming.

In a sense, serial execution could be called the ultimate in pessimism: for the duration of each transaction, each transaction holds an exclusive lock on the entire database (or a partition of the database). As compensation for the pessimism, we make each transaction execute very fast, so the “lock” is only held for a short time.

In contrast, Serializable Snapshot Isolation is an optimistic concurrency control technique. In this context, optimistic means that if there is potential danger, the transaction is not blocked — instead, it continues executing, hoping everything will turn out fine. When a transaction wants to commit, the database checks whether anything bad happened (i.e., whether isolation was violated); if so, the transaction is aborted and must be retried. Only serializable transactions are allowed to commit. If there is a lot of contention (i.e., many transactions trying to access the same objects), performance suffers because a large proportion of transactions need to be aborted. If the system is already near maximum throughput, the additional load from retried transactions can worsen performance.

Ch9: Distributed Systems
#

Clocks
#

Clocks are critically important in distributed systems — they can directly affect the visibility, isolation, and correctness of transactions. In reality, reading a precise point in time is meaningless (from a quantum theory perspective, there is no concept of an absolute point in time; the actual situation is even more complex). Spanner’s Google TrueTime API reports a confidence interval for the local clock. The confidence interval reports an extremely short and trustworthy time range rather than a time point. For example, if you have two confidence intervals, each containing the earliest and latest possible timestamps ($A = [A_{earliest}, A_{latest}]$, $B=[B_{earliest}, B_{latest}]$), and these two intervals do not overlap (i.e., $A_{earliest} < A_{latest} < B_{earliest} < B_{latest}$), then B definitely happened after A — there is no doubt. Only when the intervals overlap are we uncertain about the order in which A and B occurred. To ensure that transaction timestamps reflect causality, Spanner deliberately waits for the length of the confidence interval before committing a read-write transaction. To keep the clock uncertainty as small as possible, Google deploys a GPS receiver or atomic clock in every data center, allowing clocks to be synchronized to within about 7 milliseconds. Logical clocks are based on incrementing counters rather than oscillating quartz crystals. Logical clocks only measure the relative ordering of events.

Real time may not exist. Responsiveness trumps everything. For most server-side data processing systems, real-time guarantees are uneconomical or unsuitable. Therefore, these systems must endure pauses and clock instability in non-real-time environments.

Ch10: Consistency and Consensus
#

All the problems we’ve assumed are possible: packets in the network can be lost, reordered, duplicated, or arbitrarily delayed; clocks are at best approximate; and nodes can pause (e.g., due to garbage collection) or crash at any time.

CAP
#

The formal definition of the CAP theorem is limited to a very narrow scope — it only considers one consistency model (linearizability) and one type of fault (network partitions, or nodes that are alive but disconnected from each other). It doesn’t discuss anything about network delays, dead nodes, or other trade-offs. Therefore, despite CAP’s historical influence, it has no practical value for designing systems.

Distributed Transactions and Consensus
#

All the consensus protocols discussed so far internally use a leader in some form, but they don’t guarantee that the leader is unique. Instead, they make a weaker guarantee: the protocol defines an epoch number (called ballot number in Paxos, view number in Viewstamped Replication, and term number in Raft) and ensures that within each epoch, the leader is unique. Whenever the current leader is thought to be dead, a vote begins among the nodes to elect a new leader. This election is assigned an incrementing epoch number, so epoch numbers are totally ordered and monotonically increasing. If there is a conflict between leaders from two different epochs (perhaps because the previous leader hadn’t actually died), the leader with the higher epoch number prevails. Designing algorithms that robustly cope with unreliable networks remains an open research problem.

Ch11: Batch Processing
#

Services (online systems) Services wait for requests or instructions from clients to arrive. Upon receiving one, the service attempts to process it as quickly as possible and sends back a response. Response time is typically the primary performance metric for services, and availability is usually very important (if clients can’t reach the service, users may receive error messages).
Batch processing systems (offline systems) A batch processing system takes a large amount of input data, runs a job to process it, and produces some output data. This often takes a while (from minutes to days), so typically no user is waiting for the job to finish. Instead, batch jobs typically run periodically (e.g., once a day). The primary performance metric for batch jobs is typically throughput (the time needed to process input of a certain size). This chapter discusses batch processing.
Stream processing systems (near-real-time systems) Stream processing sits between online and offline (batch) processing, so it is sometimes called near-real-time or nearline processing. Like batch processing systems, stream processing consumes inputs and produces outputs (without needing to respond to requests). However, stream jobs operate on events shortly after they occur, whereas batch jobs wait for a fixed set of input data. This difference gives stream processing systems lower latency compared to batch processing systems.

The batch processing algorithm MapReduce, published in 2004, was (perhaps over-enthusiastically) called “the algorithm that made Google’s massive scalability possible.” MapReduce is a fairly low-level programming model.

MapReduce and Distributed File Systems
#

Compared to the query optimizer of a relational database, Unix tools, despite their simplicity, are still remarkably useful. The biggest limitation of Unix tools is that they can only run on a single machine — this is where tools like Hadoop came in. MapReduce is somewhat like Unix tools but distributed across thousands of machines. Like Unix tools, it’s fairly crude but surprisingly effective. MapReduce jobs read and write files on a distributed file system. In Hadoop’s implementation of MapReduce, this file system is called HDFS (Hadoop Distributed File System), an open-source implementation of the Google File System (GFS). Besides HDFS, there are various other distributed file systems such as GlusterFS and the Quantcast File System (QFS). Object storage services like Amazon S3, Azure Blob Storage, and OpenStack Swift are similar in many ways.

To create a MapReduce job, you need to implement two callback functions, Mapper and Reducer, which behave as follows:

Mapper The Mapper is called once on each input record. Its job is to extract key-value pairs from the input record. For each input, it can generate any number of key-value pairs (including none). It does not retain any state from one input record to the next, so each record is processed independently.
Reducer The MapReduce framework takes the key-value pairs produced by the Mapper, collects all values belonging to the same key, and iteratively calls the Reducer over this set of values. The Reducer can produce output records (e.g., the count of occurrences of the same URL).

Using the MapReduce programming model, the physical network communication aspects of computation (getting data from the right machines) are separated from the application logic (processing the data after obtaining it). This separation contrasts sharply with the typical use of databases, where requests to fetch data from the database frequently appear within application code. Because MapReduce handles all network communication, it also frees application code from worrying about partial failures, such as the crash of another node: MapReduce can transparently retry failed tasks without affecting application logic.

Another common pattern of “putting related data together” is grouping records by some key (like the GROUP BY clause in SQL). The simplest way to implement this grouping operation with MapReduce is to set up the Mapper so that the key-value pairs it generates use the desired grouping key. The partitioning and sorting process then directs all records with the same partition key to the same Reducer.

Hadoop vs. Distributed Databases
#

As we’ve seen, Hadoop is somewhat like a distributed version of Unix, where HDFS is the file system and MapReduce is a peculiar implementation of Unix processes (always running the sort utility between the Map and Reduce phases). We’ve seen how various join and grouping operations can be implemented on top of these primitives.

When the MapReduce paper was published, it was — in a sense — not new. All the processing and parallel join algorithms we discussed in earlier sections had already been implemented over a decade earlier in so-called massively parallel processing (MPP) databases. Examples include the Gamma database machine, Teradata, and Tandem NonStop SQL, which were pioneers in this area.

The biggest difference is that MPP databases focus on executing analytical SQL queries in parallel across a set of machines, whereas the combination of MapReduce and a distributed file system is more like a general-purpose operating system that can run arbitrary programs.

Diversity of Processing Models
#

Having only two processing models, SQL and MapReduce, is not enough — more diverse models are needed! And due to the openness of the Hadoop platform, implementing a whole range of approaches is feasible, something that was impossible within the monolithic MPP database paradigm. Traditionally, MPP databases met the needs of business intelligence analytics and business reporting, but this is only one of many domains that use batch processing. In the years since MapReduce became popular, execution engines for distributed batch processing have matured significantly.

Ch12: Stream Processing
#

Skipped.

Event Sourcing
#

Event sourcing is a powerful data modeling technique: from the application’s perspective, it’s more meaningful to record user actions as immutable events rather than recording the effects of those actions in a mutable database. Event sourcing is similar to the chronicle data model. Like change data capture, event sourcing involves storing all changes to application state as a log of change events. Applications using event sourcing need to pull the event log (representing the data written to the system) and transform it into application state suitable for display to users. The current state is derived from the event log.

Ch13: The Future of Data Systems
#

Lambda Architecture
#

If batch processing is used to reprocess historical data and stream processing is used for recent updates, how do we combine the two? The Lambda Architecture is one proposal for this. The core idea of the Lambda Architecture is to record incoming data by appending immutable events to an ever-growing dataset, similar to event sourcing. In the Lambda approach, the stream processor consumes events and quickly produces an approximate update to the view; the batch processor later uses the same set of events and produces a corrected version of the derived view.

Unix evolved pipelines and files that are just byte sequences, while databases evolved SQL and transactions. Which approach is better? Of course, it depends on what you want. Unix is “simple” because it’s a fairly thin wrapper around hardware resources; relational databases are “simpler” because a short declarative query can leverage a lot of powerful infrastructure (query optimization, indexes, join methods, concurrency control, replication, etc.) without requiring the query author to understand the implementation details. I interpret the NoSQL movement as a desire to apply Unix-like low-level abstractions to the domain of distributed OLTP data storage.

Separation of Application Code and State
#

In theory, a database could be a deployment environment for arbitrary application code, much like an operating system. In practice, however, they are poorly suited to this goal. They don’t meet the requirements of modern application development, such as dependency and package management, version control, rolling upgrades, evolvability, monitoring, metrics, calls to network services, and integration with external systems. I believe it makes sense to have some parts of the system specialized for persistent data storage and other parts specialized for running application code. The two can interact while remaining independent. The trend is to separate stateless application logic from state management (databases): don’t put application logic into the database, and don’t put persistent state into the application.

I assert that in most applications, integrity is far more important than timeliness. Violating timeliness may be confusing and annoying, but violating integrity can be catastrophic.

Problems Introduced by Algorithms
#

Bias and discrimination: For example, in racially segregated areas, a person’s ZIP code, or even their IP address, is a strong indicator of race. Given this, it seems absurd to believe that an algorithm can somehow take biased data as input and produce fair and unbiased output. Yet this view often seems to lurk among advocates of data-driven decision-making — an attitude satirized as “machine learning is like money laundering for bias.” Predictive analytics systems simply extrapolate from the past; if the past was discriminatory, they codify that discrimination into rules.
Responsibility and accountability: Automated decision-making raises questions about responsibility and accountability. If a person makes a mistake, they can be held accountable, and those affected by the decision can appeal. Algorithms also make mistakes, but when they do, who is responsible?
Privacy and surveillance: Let’s do a thought experiment. Try replacing the word data with surveillance and see if common phrases still sound as nice. For example: “In our surveillance-driven organization, we collect real-time surveillance streams and store them in our surveillance warehouse. Our surveillance scientists use advanced analytics and surveillance processing to gain new insights.”

Blind faith in the supremacy of data-driven decisions is not just delusional — it’s genuinely dangerous. As data-driven decision-making becomes more prevalent, we need to figure out how to make algorithms more accountable and transparent, how to avoid reinforcing existing biases, and how to fix them when they inevitably err.

Users barely know what data they’re giving us, what data goes into the database, and how the data is retained and processed — most privacy policies are ambiguous, stringing users along without coming clean. If users don’t understand what will happen to their data, they can’t give any meaningful consent. For users who disagree with surveillance, the only truly viable alternative is simply not to use the service. But this choice isn’t truly free either: if a service is so popular that it is “considered a necessity for basic social participation by most,” then expecting people to opt out is unreasonable — using it is effectively mandatory.

Summary
#

Since software and data have such an enormous impact on the world, we engineers must remember that we have a responsibility to work toward the kind of world we want: a world that respects people, that respects humanity. I hope we can work together toward that goal.

Book Notes — Average Is Over

Tue, 13 Aug 2024 00:00:00 +0000

Why I Read This Book
#

In the final pages of Elon Musk, the author briefly introduced two books by economist Tyler Cowen: The Great Stagnation and Average Is Over. The Great Stagnation is about why America’s development has stalled over the past 40 years — something I’m naturally not that interested in. But Average Is Over is not a study of history; it’s a perspective on future development, especially the impact of AI on human life.

I’ve always been interested in what human life will look like in the future. Recently, OpenAI has been hot, and it feels like the AI era is upon us. What changes will AI bring to our lives and work? Will social structures shift? Which jobs will gradually disappear? Which jobs will benefit?

Chess
#

The book spends a large portion (nearly half) discussing chess and computer programs. You can tell the author is definitely a chess enthusiast — he’s deeply knowledgeable about chess history and its evolution. Reading this section always reminds me of The Queen’s Gambit. If it weren’t for that show, I wouldn’t have known chess had rapid formats or that the Soviet Union was the world’s strongest chess nation. The author also uses chess to explore the influence of computer programs on the game.

This influence goes beyond AlphaGo defeating the world’s strongest human Go player — the “beating the brightest human minds” kind of impact. It also includes how early chess programs changed the way humans learn chess. In the early days of chess, before computers took off, people could only learn chess from other people. A beginner couldn’t often play against a chess master. But as computer programs became widespread, they were adopted en masse. Chess programs could teach you, you could play against them, and you could even set the difficulty level. This was incredibly convenient for beginners. Without us even noticing, computer programs quietly reshaped our lives. In the future, we will increasingly collaborate with AI.

Polarization
#

Once AI is widely deployed, many aspects of our lives will change. AI is unlikely to revolutionarily overturn the social structure of rich and poor; the reality that a tiny minority controls the vast majority of wealth may intensify further. The middle class is perhaps the most vulnerable stratum. Many middle-class workers perform partially intellectual but repetitive work — exactly AI’s sweet spot. The book argues that the value of middle-class work isn’t actually that great and may be relatively easily replaced. Disparities in basic assets will widen the gap in wealth accumulation — in other words, differences in starting capital will amplify differences in asset accumulation. In this age, that sentence is easy to understand.

The book approaches wealth distribution from an American perspective, but it’s easy to map onto the Chinese context. China’s economic development over recent decades has been truly remarkable — the dividends of population and infrastructure construction, a phase all developed nations went through. But the introduction of market economics and the passage of time have been accompanied by growing wealth inequality. Let’s leave it there… I don’t want to write anything too sensitive…

The Rising Cost of Learning
#

The cost of learning keeps rising. This doesn’t refer to the cost of tuition or training courses, but the difficulty of learning or mastering a profession. The word “inventor” — I suspect many people haven’t heard it in a long time. Our impression of the term is still stuck in the Edison era. Back then, individuals could invent things on their own; they just needed some relatively advanced knowledge in their field and a bit of brainpower. “Inventing” didn’t seem that hard. But as time passed, we rarely hear the word “inventor” anymore. It’s not that humans have stopped inventing — it’s that what people invent now is almost always the work of a team, many people, often requiring cross-disciplinary collaboration among multiple specialists. The cost of “inventing” things keeps rising because the knowledge required to master a field grows ever larger and more complex. It’s unrealistic for one person to master an entire industry; people tend to specialize in narrower domains — and even a narrow domain is enough for a lifetime of study.

Academia today faces this exact situation. A relatively successful paper typically requires experts from various fields to use their specialized knowledge to verify the correctness of one small segment of a proof. The book gives a classic example: if a mathematician proves a conjecture in mathematics, there may be only a handful of people in the entire world who can truly understand what the mathematician is proving. Most of them may only understand one section of the content — and even the mathematician themselves may only say: “I might be right.” We have no way to verify the correctness of the proof.

Human knowledge is becoming increasingly complex. Scientists now tend to, and increasingly do, hand calculations and experiments over to machines. Humanity seems to have reached a tipping point: our brains are nearly incapable of understanding this knowledge anymore. From a biological perspective, the human brain necessarily has a limit. The processing speed of the human brain can’t remotely keep up with machines.

Self-Learning
#

Even as learning costs rise, education will become ever more important in the future. The education system may change. Since time will be more precious in the future world and learning resources will be easier to access, people will lean toward online learning and self-directed learning. At the same time, this makes self-drive even more critical.

As an IT professional, I have a deep appreciation for self-learning. This industry is intensely competitive; if you don’t keep learning, you’re basically on the brink of obsolescence. But highly effective learning is also reflected in your salary. Our parents’ generation relied on assigned jobs and could work in one position for decades without major changes. People back then just thought about working, not obsessively self-improving and chasing certifications. Times have truly changed. How many people, like me, are still writing articles at 11 PM? I’m even baffled by industries where you don’t need to keep learning after graduation — just how backward are they? You graduate university in your early twenties and still have decades to learn. It would be utterly strange to just stagnate there. Of course, I don’t like cutthroat competition, but I like standing still even less — especially in an age where just sitting on a stool spacing out causes the wealth gap to widen.

Finally
#

The cover and illustrations for this post are all AI-generated. I just typed in “goodbye age of mediocrity” and the AI produced astonishing images. I don’t know exactly which industries or professions will disappear in the future, but at the very least, illustrators are going to have a hard time surviving in the AI era.

AI has already invaded the IT domain. As a DBA, which of our work patterns will be replaced? That’s a question worth pondering. Whatever happens, in this age, only learning can keep you competitive. I hope none of us will be the “disappearing shoulder pole porter.”

Book Notes — Educated and Atomic Habits

Tue, 13 Aug 2024 00:00:00 +0000

I’ve actually wanted to write about these two books for a long time. I love reading, but I absolutely detest writing. Maintaining a blog is practically a miracle for me. I love reading because I love and believe in the power of education. As for how much I hate writing, let me tell you a little story.

I Hate Writing Essays
#

My dislike for writing is practically innate. Since elementary school, I never wrote diaries or essays. Every winter and summer break homework required a daily diary entry — I never wrote a single one. I still remember when school started and I had to turn in summer homework. The teacher threatened that if I didn’t finish it, I couldn’t attend class. I still wrote nothing and just sat in the classroom as usual. Later, for some assignment, our homeroom teacher had to submit examples of correcting typically flawed students — only 4 or 5 kids in the class were selected. One was bad at sports, one had a temper problem… and I was the one bad at writing! And the task was to write an essay about correcting that flaw! I can’t write — why would you make me write an essay about fixing my inability to write??? I dragged it out for two weeks. All the other students turned theirs in. I couldn’t squeeze out a single word. The homeroom teacher personally coached me. She said: when you walk down the street, you can turn anything you see into a sentence. See a blue sky? You can form a sentence in your mind: “The sky is cloudless for miles.” Practicing sentence construction regularly can help with writing. You could also think of other ways to fix your writing aversion. Another week passed, and I wrote down exactly what she told me, verbatim. I could see the frustration in her eyes. Later, in middle school, I cleverly befriended the Chinese class representative so she’d leave my name off the missing-homework list. That’s how I dodged three years of middle school. Then in high school, I once awkwardly wrote an essay my own way and scored 30 out of 60 — a devastating blow. So for every monthly exam, I simply didn’t write the essay. I figured: it’s just mock exams, not the Gaokao — I’ll just forfeit those 60 points. Finally, for the actual Gaokao and the two mock exams before it, I crammed Qu Yuan and Li Bai into essay templates like eight-legged essays. I found there was nothing you couldn’t cram them into, and I muddled through the Gaokao essay hurdle. College? No need to mention it — my hand had forgotten how to hold a pen.

Yes, with this peculiar writing psychology, I hated essays. But after entering the workforce, I gradually understood: the dullest pencil is better than the sharpest memory. No matter how many books you read, you need to internalize them. The pressures of ambition, family, and work forced me to change. Whether I needed this skill or not didn’t seem to matter — if society needs it, I should try to adapt. Writing not only pushes you forward, it’s also a way to record growth, to record life. Even my own technical articles — years later, I still have to come back and read them carefully, review them carefully.

Reading Originals Is Good for Body and Mind
#

I read both Educated and Atomic Habits in their original English. I had a bit of an English foundation, and since I was preparing for graduate school entrance exams at the time, I wanted to improve my English reading — so I chose English originals. At first, reading English originals was quite difficult. Many words were unfamiliar, and I’d look them up and annotate them in the book. Progress was painfully slow. But as I read deeper, there were fewer and fewer annotations. It wasn’t that I quickly memorized many new words — rather, some important words appear repeatedly throughout a book, while others that appear rarely don’t affect comprehension. Also, at the start you don’t know what the book is about, so comprehension latency is high. Later, once you know where it’s headed, reading naturally speeds up. For instance, the word “ridge” appeared very frequently early on, and I eventually remembered it. Some similar words I still can’t remember, but I know they’re some kind of geographic term — summit, valley, ridge — and even without remembering them precisely, it doesn’t stop me from reading. That’s how English originals work: difficult at first, faster the further you go.

Educated
#

The Chinese title of Educated is You Should Fly Like a Bird to Your Mountain — I really want to complain about this title. “Educated” is the spiritual essence of the entire book, and near the end, the author drives it home with “call it educated” — absolutely brilliant. This Chinese title is like shit, completely missing the book’s essence. The author’s personal experience is legendary: a child who walked out of some corner of the American mountains, who through sheer effort studied all the way to Cambridge. Her father was uneducated, anti-social, lacking basic physics knowledge — his ignorance led to family members getting injured or even disabled. He disapproved of children going to school, even believing education was government brainwashing. Countless absurd behaviors. Her brother also had personality issues — he shoved her head into a filthy toilet and made her beg for mercy, then the next day acted like nothing happened and continued being her “good brother”… Later, the author found her way out through education, and in the end, she didn’t want to return to that valley. Reading the ending always reminds me of my own experience. Of course, I didn’t have such an extreme environment, nor such a legendary journey, but I feel like I can understand — after being educated, family interactions somehow feel unnatural. It’s not about getting cocky after university — the generation gap is real. I deeply believe in the importance of education. If my family hadn’t sold everything they had to fully support my education, our circumstances would never have changed. If you’ve truly been mired in poverty, you know how fierce the desire to escape it is — and education is almost the only way out for people like us. Educated is a great book: clear prose, comfortable sentence structure, suited to modern reading rhythms, a gripping story, a profound theme. It’s an excellent choice as your first English original.

Atomic Habits
#

Atomic Habits — I’ve forgotten exactly how I found this book, but it changed my understanding of behavior. Building good habits isn’t actually that hard; most people just don’t know how. Many have said: I’ll read X books in a few months, run Y kilometers, lose Z pounds — but they rarely follow through. Building good habits requires genuinely liking the habit, changing your mindset, reducing the friction of the action, putting obstacles farther away, forming reward mechanisms, and so on. When you want to become a certain kind of person, don’t focus on how to become that person — think about what that kind of person does, and learn to do it. For example, quitting smoking: if your brain thinks you’re “in the process of quitting,” it’s very hard. If someone offers you a cigarette and you say “I’m quitting,” a few words from them might get you to smoke. But if you genuinely believe you’re someone who “doesn’t smoke” — note, this must be your authentic inner belief — when someone offers you a cigarette, you’ll simply say “I don’t smoke,” and you probably won’t have to smoke it. Some small details: say you want to build a habit of reading at night — you need to break the habit of scrolling on your phone. Move your books from the bookshelf to your bedside for easier access. Put your phone at the foot of the bed, making getting up the barrier to grabbing the phone — this makes it easier to reach for the book instead of the phone. If picking up the book is still hard, reframe your thinking: “reading” as an action may feel difficult, but break it down — “pick up the book” or “open to the first page” becomes your mental target. The startup action for reading is simple and easy to complete. After reading the first page, think about what comes next — and in reality, once you’ve read the first page, it’s hard not to read the second. Of course, there are many more excellent suggestions for building good habits and shedding bad ones — every word is a gem, thoroughly engaging. After reading Atomic Habits, whenever I want a certain habit, I first consider the book’s guidance, then plan how to implement it — rather than acting on impulse.

Finally
#

At last — these two books have had an enormous impact on me. One is a legendary autobiography; the other is a behavior-transforming book. Neither is the kind of work you forget shortly after reading. They’re perfect starter books for cultivating a reading habit, especially for those wanting to read English originals. I really don’t recommend Pride and Prejudice or One Hundred Years of Solitude — yes, they’re classics, but their impact on the reader is quite low, and they were written so long ago that some vocabulary and grammar are too archaic, making them unsuitable for first-time English readers. Looking at this through the lens of Atomic Habits: reading these English classics is not only more difficult but also lacks immediate personal benefit, making it hard to form a habit.

Book Notes — Elon Musk

Tue, 13 Aug 2024 00:00:00 +0000

Gifted
#

Musk’s ancestors, driven by a love of adventure, emigrated from America to South Africa. His maternal grandfather even flew a plane from Africa to Australia. Musk was born in South Africa and showed astonishing memory and brilliance from an early age. His mother, Maye Musk, told his teacher: “My son is a genius.” The teacher replied, “Yes, every mother says that.” Maye: “No, I mean he really is a genius.” As a child, Musk sometimes seemed “slow to react.” His mother said when people talked to him, he’d give no response at all. She thought something was wrong with his brain and even took him to a doctor. But later she discovered Musk was simply immersed in his own world of thought. As a child, Musk could even finish reading the entire library’s collection and then ask the library to get more books…

Arriving in America
#

Due to the less-than-ideal environment in South Africa, Musk, approaching university age, executed a two-step jump. He first went to university in Canada, then to the United States for his master’s. Upon finally reaching America, Musk immersed himself in Silicon Valley’s work environment. The tech industry desperately needed young people like him — brilliant and relentless. And Silicon Valley’s tech atmosphere and culture of freely exercising one’s talents let Musk dive in completely.

Zip2 and PayPal
#

Soon Musk founded Zip2, essentially a corporate version of online maps. While we’re now very familiar with online maps, the US internet industry was just getting started back then — this was all novel stuff. After many twists and turns, Zip2 did grow. Personally, I think Zip2’s model would have struggled to survive long-term without pivoting toward online maps or something like Yelp. Eventually, some sucker bought Zip2 for $300 million, instantly turning Musk into a multimillionaire and Silicon Valley tech tycoon. You could actually tell — Zip2 was deeply divided internally, had directional problems, and Musk didn’t have absolute decision-making power. He probably wanted out long ago.

Before leaving Zip2, Musk was already planning and recruiting for online payments. At that time, the world didn’t even have anything like Alipay… Musk believed traditional finance was too conservative and that there was enormous opportunity to change the industry model. But many bankers didn’t believe internet finance could work, because internet finance couldn’t handle network security issues — after all, the slightest error in finance could have enormous consequences. Initially, the company Musk founded wasn’t PayPal but X.com, which later merged with PayPal and kept the latter’s name. Early on, X.com suffered massive attacks but survived. Their security mechanisms at the time had a significant influence on the later online payments industry. PayPal was later acquired by eBay, netting Musk hundreds of millions of dollars — another huge payday.

SpaceX and Tesla
#

Zip2 and PayPal were, for Musk, validation of his industry sensitivity and business acumen — though some questioned his execution and decision-making abilities, i.e., his CEO chops. As always, Musk viewed these industries as too conservative and old-fashioned. Musk loved recruiting extremely capable top university graduates and disliked hiring seasoned, conservative-minded industry veterans. He ran both companies simultaneously, and for a long time, neither company produced any product at all. And, as you’d imagine, rocket-building burns through money like nothing else. After several failed rocket launches, Musk deployed his signature skill: fire… And just as the financial crisis hit and no one wanted to invest, he poured his entire personal fortune into both companies. After several failures, SpaceX’s Falcon rocket finally achieved the feat of being the first private company to successfully launch a satellite, landing a $1 billion NASA contract. Tesla, after shamelessly asking early Roadster customers for more money (because developing such a radically new-concept EV cost far more than projected), finally produced a finished vehicle and built out a highway EV charging network and an electric car factory. After simultaneously succeeding with two industry-disrupting companies, no one questioned Musk’s ability anymore.

For Humanity
#

Musk’s success is inseparable from his excellent qualities: sensitivity to future technology, rapid comprehension of new industries, talent identification, a free and open tech and market environment, long working hours and execution… But the things people dislike include his ruthlessness toward employees — some loyal, devoted people, fired just like that. As a worker myself, I deeply understand the feeling of giving your all without recognition from the company. Reading this book, I could even feel how American capitalists truly exploit workers. Once, an employee missed a company gathering because he didn’t want to miss his daughter’s birth. Musk emailed him an angry tirade: do you want to wallow in domestic trivialities or work relentlessly to change the world? The guy just didn’t want to miss his daughter’s birth.

A few years ago, reading Steve Jobs, I thought: how could someone be so obsessive? But that exact kind of person changed the mobile industry and brought about the smartphone revolution. Jobs was way too formidable. After reading Elon Musk, I now feel Musk is even stronger than Jobs. Tesla, SpaceX, SolarCity — they’re all oriented toward humanity’s future. The future world seems to have started its engines; you can see it slowly arriving.

Musk’s Mars plan finally seems to have glimpsed some dawn. For decades, the American space industry had nearly stagnated. He brought a new model and once again made aerospace a hot field. But there are also uncertainties. If a crewed launch explodes and causes casualties, SpaceX could plunge back into the abyss. And if Tesla discovers a serious defect requiring a mass recall, the stock price would crash.

If you could be the first human to set foot on Mars, would you do it? Musk has thought about it, and he truly could become that person. But Musk wouldn’t do it. The book’s original words: I want to go, but I don’t have to. The point is to enable many people to go to Mars. It would be like the head of Boeing being a test pilot — for space exploration, that’s unwise. Even never going to space is fine. The point is to extend the lifespan of humanity as much as possible.

Working for humanity — this theme truly stirs the heart. I’ve played Civilization VI for days and nights on end, from stick-wielding primitives to igniting rockets, all for that moment of launch, when humanity becomes an interplanetary species and builds a new home on Mars!

Book Notes — Rich Dad Poor Dad

Tue, 13 Aug 2024 00:00:00 +0000

Rich Dad Poor Dad. I used to scoff at this kind of book. It’s the type of success-literature you see displayed at bookstore entrances, looking insubstantial at a glance, very unreliable — the kind of thing that seems to prey on people at the bottom of society who dream of getting rich quick but can never actually apply the book’s advice due to their own circumstances or environment. Besides, smart people don’t read such uncultured books, right? The title is tacky as hell!

I spent a period watching Lao Gao and Xiao Mo on Bilibili, and one episode talked about this book, making it sound mystical and mysterious. Plus, it’s a global bestseller, so I bought it to see just how magical it really was.

Started Reading E-Books
#

I’m a die-hard fan of paper books. I love the feeling of finishing an entire book and placing it on the shelf to collect — that “this whole bookshelf is my knowledge” feeling. I wasn’t really into e-books; they give a “read it and it’s gone” vibe. Three reasons brought me back to e-books:

I recently deleted all my go-to time-killing apps and needed an app that wasn’t so brain-numbing — something I could open first when pulling out my phone.
E-books are just more convenient than paper books; you can pull them out and read anytime.

Making use of fragmented subway commute time. Back when I was preparing for graduate school entrance exams, I made a detailed daily schedule that included subway time. Since I’d already built the habit of studying on the subway, I didn’t want to give it up. I recommend my summary of the graduate exam experience: How I Got Into Wuhan University’s Part-Time Graduate Program

I slightly adjusted my old plan — no need to grind vocabulary as intensely anymore — so I swapped in e-book reading. I split subway time into two blocks: morning commute and evening commute.

In the morning, my mind is clear and my mental state is good, so I read technical e-books. These require slow reading, sometimes stopping to think. This is goal-driven reading.

In the evening, my mind is foggy (not really foggy, more often it’s a headache), so I read lighter books — like “extracurricular” books such as Rich Dad Poor Dad. These books aren’t hardcore in content, so I read fast and enjoyably, with a bit of a dopamine-driven reading feel.

Poor Dad and Rich Dad
#

Author Robert Kiyosaki grew up in Hawaii. His biological father was a highly educated government education official — the “Poor Dad.” His best friend’s father was a high school dropout with extraordinary financial intelligence — the “Rich Dad.” Poor Dad had higher education but worried daily about loans and bills, while Rich Dad spent every day directing people to create wealth for him. The book has a classic line: “The poor work for money; the rich make money work for them.”

The mindset of making money: As a child, the author and his good friend came up with many ways to earn money. They once used toothpaste tubes to counterfeit coins — not knowing at the time that making money was illegal — and were stopped by adults. Later, they gathered free books from stores, set up a little library in a spot, and earned money by renting books to neighborhood kids. They stopped after attracting some local unsavory characters. Rich Dad admired their money-making behavior. He believed the difference between the rich and the poor is that the rich are always thinking about how to make money, while the poor are only thinking about how to find a good job.
On taxes: The poor pay far more in taxes than the rich. Rich Dad paid more taxes than Poor Dad, but Rich Dad’s income was vastly higher. When the US president decided to raise taxes on the rich, they only raised taxes on the salaried middle class — the truly rich were unaffected. The rich have many ways to legally avoid taxes, by understanding how to use the law. For example, the author says in real estate: if you sell a house, the income is heavily taxed, but if you swap houses, there’s no tax. The rich can use this statute to invest in real estate and legally avoid taxes. But the poor can’t escape income tax — the more you earn, the more tax you pay.
On investing: Investing requires cultivating knowledge in accounting, finance, and law. In short, if you want to make money, you need to develop your financial intelligence. When you earn a big sum, you should start the next investment rather than buying consumer goods. Think of the tables, chairs, jars, bottles, clothes, and household items in our homes — we pay a relatively high price for them, but the moment they’re bought, their value drops to near zero. This isn’t to say don’t buy things — but consider investing your money first, then consider zero-return consumption.

Finally
#

The book says our education system cultivates people’s ability to work, not their ability to make money. I strongly agree with this statement, but I still believe in the power of education. The author isn’t telling people to skip education — education is also very important for making money. We need to understand the basic operating principles and rules of this world, and that can help us find suitable ways to make money.

Given the author’s family background at the time, they may have been relatively poor compared to Rich Dad, but for truly poor people, their family conditions were far from poor. I feel my own circumstances haven’t yet reached the point where I can fully devote myself to investment and money-making. If one day I have some spare money and my management, social, and decision-making skills reach a certain level, I might look for ways to make money. For now, I can’t think that far — gotta code well and fill the holes first.

The Learning Pyramid in the book left a deep impression on me. You really retain very little from passive learning. That’s why I persist in writing frequently, including book notes like these. Problems I encountered during years of late-night database maintenance — I still remember them vividly. Hands-on experience truly creates the deepest memories. That said, hands-on experience is unpredictable and rare; reading and self-learning are the lowest-cost, easiest-to-form habits, and the most cost-effective way to improve ability. They’re not that “passive.” The Learning Pyramid’s “passive” refers to knowledge being received by the subject; “active” refers to the subject outputting knowledge. This also strengthens my motivation to record and share — whether technical or non-technical.

Book Notes — Sapiens: A Brief History of Humankind

Tue, 13 Aug 2024 00:00:00 +0000

This is a book I spent a long time reading. It’s thick, covers an enormous range of topics, and tackling the original English edition was challenging. But thankfully, I finally finished it — today (February 2023). A real sense of accomplishment.

Sapiens: A Brief History of Humankind is a grand history book that comprehensively introduces the development of human civilization. I’ve always enjoyed learning about human history, immersing myself in its weight and the vitality of civilizational progress.

The Cognitive Revolution and Fiction
#

Conventional views of human history hold that humanity’s first major evolution or revolution was learning to use tools. Like in 2001: A Space Odyssey, where apes bang bones together as the iconic BGM plays — but that’s science fiction. Sapiens argues that humanity’s first major revolution was the Cognitive Revolution, the key distinction between humans and animals. Learning to walk upright didn’t just free our hands — more importantly, it freed our minds. Four-legged running animals never evolved the way we did because harsh natural environments demanded stronger bodies and limbs for speed. Walking upright obviously makes you slower, so group living and tool use compensated. But group living and tools aren’t unique to Sapiens — many animals live in groups, and chimpanzees use tools too. What set Sapiens apart was learning to manufacture weapons, boats, and sustain much larger social groups. They walked from Africa to the Middle East, to Europe, battling the physically stronger Neanderthals and ultimately taking their territory. They reached the Far East, crossed the Bering Strait into the Americas, and even sailed to Australia. This ability to craft complex tools and communicate at unprecedented levels — that’s what the Cognitive Revolution brought.

Neanderthals themselves have gone extinct, but recent research shows the vast majority of humans carry a small amount of Neanderthal DNA — except for indigenous Africans. This suggests Neanderthals weren’t entirely wiped out by Sapiens; a small number interbred with Sapiens and their genes spread across the world. This is also key evidence supporting the Out-of-Africa theory of human origins.

The book gives a classic example of the Cognitive Revolution: imagine a lion by the river. One Sapiens sees it and tells others. The others then construct in their minds the idea that “there is a lion by the river” — even though they don’t know for certain whether one is actually there. The prerequisite is that Sapiens had to learn to conceive of things that aren’t immediately present. More importantly, once they mastered this skill, language, fiction, lies, power, social structures followed… Neanderthals clearly exchanged far less information than Sapiens.

The Cognitive Revolution had an enormous impact on civilizational development. It allowed the construction of things that don’t actually exist — gods, religions, power, money, social structures, dynasties… Take a company, for example. A company is really a social construct; it doesn’t actually exist in the physical world. A company can be a stack of 4A paper with a stamp in a document bag — but that’s just paper. Employees believe the company exists because their minds believe it does. Everyone believes it exists, but the company itself is a fiction in human minds — the entity “company” does not exist in the real world.

Money
#

How did money come about? In a world without money, stable social structures gave rise to barter trade. But as the variety of traded goods increased, the number of equivalent exchange equations grew exponentially. When trading shoes for a rake, it’s a simple one-for-one swap. Add a donkey, and you have three exchange equations: shoes-rake, rake-donkey, shoes-donkey. As goods multiply, the number of exchange equations becomes a combinatorial explosion — and that’s not even accounting for multi-item trades. Then an intermediary — money — appeared and solved the problem instantly. Everything only needed to be equated with money. Money served as the universal equivalent for all goods, and the convenience of trade improved dramatically. Early forms of money were diverse, with shells being the most common. If shells were too easy to obtain, someone could buy up everything in the market, so shell-based monetary civilizations were typically inland. Since people carried money in their pockets to buy things or hoarded it at home, and worried that too-easy acquisition of currency would disrupt markets, gold — rare, resistant to decay, difficult to mine — became humanity’s primary currency for long periods. In ancient Europe, many kings minted gold coins bearing their portraits or logos, resulting in a vast variety of European gold coins. Ancient China was somewhat different: starting with shells unearthed at Sanxingdui, then bronze coins during the Spring and Autumn and Warring States periods, then gold, silver, copper, and paper money (jiaozi) across dynasties. China didn’t stick to gold like Europe did, mainly because the population was too large and gold reserves too small, making gold too valuable — they needed other metals to create a monetary gradient to smooth trade across different scales.

There’s another great insight in the book: money and religion both have a certain transmissibility. Money and religion are essentially no different — they are both human constructs, fictions. Their only difference: religion tells you what you should believe, while money tells you what others believe.

Columbus and Zheng He
#

The Age of Discovery, the early Industrial Revolution. Europeans were passionate about exploring the world’s unknown territories. After Europeans learned the Earth was round, lacking good surveying tools, Columbus set sail westward from Europe aiming for India. They crossed the Atlantic and reached a landmass, encountered the locals, thought they’d reached India, and called them “Indians.” To this day, “Indian” in the United States carries both meanings. Europeans realized the world still had many corners untouched (at least by relatively modern civilization). They redrew world maps, filling unknown regions with sea monsters and leviathans. These maps are still widely used in video games — for instance, Civilization VI uses sea monster maps for unexplored territory, waiting to be discovered. Europeans eagerly sought new lands, and soon South America, New Zealand, Australia, and countless small islands were discovered and claimed. Where local civilizations were too far behind — the Aztec, Native American, Māori, Tasmanian civilizations — they were brutally massacred, their lands occupied by white settlers.

When the Aztec civilization encountered Spaniards clad in gleaming iron armor and wielding sharp iron swords, they thought those men were gods. They couldn’t comprehend such hard clothing and weapons — they must have been sent by the gods. And then they were deceived and slaughtered by “higher civilization.”

Zheng He’s ships were called “dragon boats” (the original text says this, even includes illustrations — I think it may be a mistake, or Westerners assumed any ship with a dragon figurehead was a dragon boat). They were several times larger than Columbus’s ships and set sail one to two centuries earlier. Zheng He’s fleet, with far superior technology, discovered new lands but didn’t occupy them — they traded with the locals. The book argues that Europeans were more adventurous and aggressive, thus ushering in the Age of Discovery. It seems Europeans hold a relatively friendly view of the Ming Dynasty. In Civilization VI, only three Chinese leaders appear: Qin Shi Huang, Wu Zetian, and Zhu Di — and only Zhu Di of the Ming Dynasty is the “tall build” development-focused leader.

Closing
#

Retracing the development of human civilization lets us understand where we came from, what we’re doing now, and explore where we’re headed. This love for the subject is also why I enjoy strategy games like Civilization VI and Humankind. When you plant rice, domesticate horses, mine salt, iron, coal, oil, uranium… there’s a thrill of human progress.

I’d like to close with a quote from Civilization VI, a game I’ve played for over 400 hours: “From the first stirrings of life beneath the water… to the great beasts of the Stone Age… to man taking his first upright steps, you have come far. Now begins your greatest quest.”

Book Notes — 2001: A Space Odyssey

Mon, 12 Aug 2024 00:00:00 +0000

 Arthur C. Clarke's masterpiece — a work no sci-fi fan can afford to skip. I'd long heard of its reputation, but having already seen the film adaptation, I felt it lacked some novelty, so the book just sat on my shelf unread. But after reading it, I can say with complete confidence: every page is filled with freshness — the kind of dopamine-driven reading that makes it impossible to put down.

God-Tier Predictions
#

This book was published in the 1960s — more than 60 years ago from now (2023). What is science fiction? Sci-fi makes reasonably plausible predictions about the future based on current science. And the author, living in the 1960s, imagined humanity’s space exploration in the year 2000. We, living in the present, are perfectly positioned to verify his “future world.”

Of course, these prophetic predictions aren’t perfectly accurate. For example, his forecast for manned space travel is clearly a bit too optimistic. After the Apollo program ended, we’ve never again undertaken a practice that breaks free from Earth’s bounds — not even returning to the Moon…

In the novel’s year 2000, humanity already has a luxurious Moon base and dispatches astronauts aboard a spacecraft bound for Jupiter.

But you can’t really blame the author. The book was published in 1968, and the very next year, humans landed on the Moon. Given a few more decades, landing on Jupiter should’ve been feasible, right?

The novel contains many astonishing predictions — here are a few that left a deep impression:

Population: Arthur C. Clarke predicted with stunning accuracy that the global population would explode to 6 billion by 2000 (in the 1960s it was 3 billion). He even foresaw certain countries implementing birth control due to overpopulation, limiting families to two children. (Clearly conservative, right… The Celestial Empire had already started family planning, and only one child was allowed — until young people stopped wanting children altogether.)

Pandemic control: In the year 2000, a global pandemic spreads, with quarantine zones set up everywhere… (I have no f***ing words.)

Artificial intelligence: In 1946, von Neumann invented the computer — the concept was just emerging — yet Arthur C. Clarke was already emphasizing the concept of artificial intelligence, predicting AI’s control over vast, complex systems. Even more remarkably, he had already imagined AI potentially rebelling against humans… ChatGPT was only recognized this year. The more you think about it, the more chilling it gets~

Tablet computers: Home computers didn’t appear until the 1980s, yet in the novel, people are already using tablet computers to control system inputs and read the news… Because the novel is so hardcore, Clarke even describes switching between a news homepage and category pages on a tablet, with data analysis delivering content tailored to the user…

Triple-site mirroring: As a DBA, I’m hyper-sensitive to this term. The author describes data center mirror backups, with data split into three identical copies stored in different locations on Earth for disaster recovery… I’m not entirely sure when concepts like “two-site-three-center” or “three-site-five-center” were first proposed (though I imagine not long ago), but seeing the novel describe data mirroring and remote disaster recovery in such detail genuinely struck a chord with my DBA instincts.

Reading this masterpiece, my state of mind was: shock, then more shock, then nonstop shock~ How did Arthur C. Clarke, in the 1960s, conceive of this future world? Unimaginable. No wonder some people say: “Arthur C. Clarke time-traveled to the present, then went back to the 1960s to write this work.”

Imagination
#

 If it were merely scientific prediction, it couldn't truly be called science fiction. Sci-fi can't just be cold scientific extrapolation — it needs a touch of humanistic distillation, a bit of imagination that departs from science, like Liu Cixin's portrayals of human nature. This element of imagination beyond science is precisely what ultimately determines a sci-fi work's stature.

 And the ultimate imaginative conceit of *2001: A Space Odyssey* is the TMA-1 monolith and the Star Child. The TMA-1 monolith is an alien artifact that catalyzes human evolution, and it simultaneously represents the vast gap between human science and alien science. The entire novel revolves around this monolith — it is the very core of the entire sci-fi story. In fact, the monolith only appears at two points in time: the ape-men era and the beginning of humanity's space exploration. When the ape-men first encounter the monolith, their physical structure undergoes subtle changes — their hands become more dexterous, their brains begin to think. The author then uses several chapters to describe the ape-men's transformation:

 1. This group of ape-men masters tools. In a confrontation with a leopard, for the first time in history, they gain the upper hand — marking the first time they stand at the top of the food chain, no longer prey.

 2. This group of ape-men decisively triumphs in a struggle against another group of apes — marking their transformation from ape-men into humans.

 Then, the novel leaps over millions of years of human history, cutting directly to the era of space travel. This technique is utterly brilliant~

 The second time: a lone human, after countless hardships, reaches the monolith on Saturn (Jupiter in the film). The protagonist passes through a wormhole pre-arranged by the alien beings, experiences a journey through space, witnesses many wondrous cosmic spectacles, and finally falls into a room — the Star Child is born!

 The alien life guided ape-men to become humans, then guided humans to become the Star Child. The Star Child is pure imagination — built on the analogy of ape-men becoming humans, marked by the TMA-1 monolith. Imaginative elements are added perfectly and naturally, leaving a profound, lingering aftertaste. Worthy of being a seminal work in science fiction.

Old Liu (Liu Cixin)
#

 I read quite a few of Liu Cixin's works during university — *The Three-Body Problem*, *The Wandering Earth*, *Ball Lightning*, *Earth Cannon*... I really like *The Three-Body Problem*, but I have no interest in the excessive factional disputes in the first book — I even found them a bit contrived. However, the concept of understanding Trisolaran society through the Three-Body game is brilliantly executed. *The Dark Forest* is clearly much better — arguably the most thrilling book in the trilogy. Back when I finished these works, I had a feeling *The Wandering Earth* might be adapted into a film; the others seemed harder to film...

 Liu Cixin's sci-fi works feature strong narrative suspense and abundant human conflict, focusing more on human behavior against a cosmic backdrop. Arthur C. Clarke's works, by contrast, rarely dwell on interpersonal relationships. He prefers depicting the face of future society and the bizarre wonders of stars, planets, and space travel.

 Many parts of Liu Cixin's work clearly show the influence of *Space Odyssey*. When Clarke describes TMA-1, he uses the word "smooth" — clearly the "droplet" in *The Three-Body Problem* references this concept. Both are technological products of alien civilizations beyond human comprehension, though their purposes are vastly different~

 Speaking of which, Liu Cixin hasn't released a new work in over a decade — what's he up to...

The Film — 2001: A Space Odyssey
#

 Released in 1968, another masterpiece by Kubrick — the god of sci-fi meets the god of cinema.

 That iconic BGM swells~ When the ape-man throws the bone — the tool — into the sky, and as it falls, the shot cuts to millions of years later... An exquisitely brilliant piece of cinematic language, truly stirring~

 When I first watched this film, there were many parts I didn't fully understand. After reading the novel, everything falls into place. The film also adds many classic scenes, such as:

 1. The depiction of Earth's orbital space in the year 2000. After over 30 years of development, humanity has launched countless capsules into space — the sky is filled with all manner of spacecraft. This sequence was frequently referenced before the year 2000.

 2. HAL 9000 reading the astronauts' lips and learning they plan to shut him down. I assumed this scene was in the novel, but the book's portrayal of taking down the AI is far more circuitous. Both are brilliant, though. (The film *The Wandering Earth*'s MOSS pays heavy homage to HAL 9000.)

Closing
#

 *Space Odyssey* perfectly embodies what hard sci-fi should be: god-tier predictions about the future, paired with a finishing touch of pure imagination. I read this book far too late — I absolutely must read the sequels soon!

Book Notes — Are We Smart Enough to Know How Smart Animals Are?

Mon, 12 Aug 2024 00:00:00 +0000

Preface
#

My previous book was Wild — the Pacific Crest Trail queen mentioned this book, Are We Smart Enough to Know How Smart Animals Are?, noting how she’d read it page by page, tearing each one out after reading. I wonder: as she journeyed through mountains and forests, hearing birdsong and streams, did reading this book about how clever animals are feel especially resonant?

I’d previously read Sapiens (I can’t help recommending this book — it’s incredible). That book starts from when humans first stood upright and traces our journey until we gradually became gods… What exactly makes humans different — what allows us to stand out from the myriad of living creatures?

The author, Frans de Waal, is an expert in primate behavior — the most cutting-edge and popular field within all animal behavior studies. Especially as experimental methods have improved, we’ve discovered that those traits humans keep proudly claiming as uniquely ours have all been found in other animal groups.

Apes
#

This book is highly scientific, containing extensive descriptions of experiments, observations, and the development of biological science. Since it’s science, let’s learn something~ When you see the word “ape,” what kind of ape image comes to mind? Whatever it is, it’s not precise enough. Because “ape” is a general term — you can roughly divide apes into four types: chimpanzees, gorillas, orangutans, and gibbons (bonobos are likely a branch of chimpanzees, frequently mentioned in the book; I’ll set them aside for simplicity):

Homo Sapiens	Chimpanzee	Gorilla	Orangutan	Gibbon
Homo Sapien	chimpanzee	gorilla	orangutan	hylobates
/>	/>	/>		/>

Kinship:

Hominoidea means the family of “hominoids” — and yes, all these close relatives of ours belong to the hominid family! The other Homixxx entries are smaller tribal branches. From the family tree above, we can see that we Homo sapiens are most closely related to chimpanzees, with gorillas, orangutans, and gibbons increasingly distant.

Evolutionary timeline:

About six million years ago, we and chimpanzees were still the same species… Chimpanzees are also universally recognized as the most intelligent animals. Did we really evolve from monkeys? This description isn’t quite accurate. Although the diagram above doesn’t mark monkeys, going further back we certainly share a common ancestor. But that doesn’t mean we evolved from monkeys — just like chimpanzees, we share a common ancestor that is now extinct. So we didn’t evolve from monkeys, but we and monkeys share a common ancestor — just two different branches. “Although for convenience we often use ‘animals’ to refer to non-human species, it’s undeniable that humans are a kind of animal.”

What Makes Us Different?
#

Tool use?

After reading Space Odyssey, I thought what made humans human was our learning to use tools. From the moment we grasped tools in our hands to crack open bone marrow, to humanity venturing into space to explore the unknown — all because we learned to use tools. But we can easily find similar behaviors in other animals. Chimpanzees use twigs to eat ants, and use branches as ladders to climb over walls. Even their thumbs, like ours, can grasp objects. Tool use is actually quite common in the animal kingdom. It seems tool use is not a uniquely human trait — those animals that also possess this skill haven’t developed higher civilizations.

The Cognitive Revolution?

After reading Sapiens, there was one particularly novel idea. I long and firmly believed it was correct: the Cognitive Revolution. The author argues that the Cognitive Revolution was the crucial juncture where Homo sapiens diverged dramatically from other animals. The Cognitive Revolution occurred before the Agricultural Revolution, when sapiens were still just hunters. The author gives a classic example: one person discovers a lion by the river, and returns to tell the rest of the tribe — “There’s a lion by the river.” At that moment, even though no one else has seen it with their own eyes, they all believe in their minds the concept of “there’s a lion by the river.” This transmission of belief later gave rise to religion, power, nations, currency, corporations, and other virtual concepts. Are We Smart Enough to Know How Smart Animals Are? offers a counterexample: a monkey, being bullied by two others, cornered with no escape, lets out a “snake!” cry (the call they only make when they encounter snakes). The two other monkeys stop to check whether there really is a snake — only when they confirm there isn’t one do they resume the chase. Many observations show that numerous animals possess the ability to believe through others’ stories.

Upright walking?

Upright walking freed our hands, and our brains grew increasingly developed. This is described in Sapiens. In fact, bipedal walking isn’t as special as we imagine. Bonobos on the savannah can walk on two legs for extended periods.

Language?

Language was once thought to belong to humans alone. Just because we can’t understand what animals are saying doesn’t mean they lack simple language. Animals’ various calls are not innate. When a chimpanzee grows up with one group, their calls in different situations are similar. If you place that chimpanzee in a different, unrelated chimpanzee group, researchers found their calls are completely different — and for a long time, that chimpanzee cannot integrate into the new group until it learns the new calling patterns. Some once believed language influences how we think. But to think, language is not a necessity. The ability of animals to add different numbers was once thought to depend on language, yet in an experiment, a chimpanzee successfully added numbers.

Cooperation?

The Wandering Earth 2 has this scene: a minister shows a fossilized human bone that was broken and healed — proof that this human suffered a severe injury. Among other animals, the injured would be abandoned, but this person received help from others and survived. Is cooperation the dividing line between humans and animals? Chimpanzee groups help elderly chimpanzees with limited mobility — bringing them food, feeding them water mouth-to-mouth.

Complex social relationships?

Chimpanzees not only know their own relationship with other chimpanzees, but also understand the relationships between B and C. Even when encountering an unfamiliar chimpanzee, they can assess its social status through how other chimpanzees treat it, and behave accordingly.

Thinking about the future?

Absolutely no problem at all…

Plato proposed that humans are the only featherless bipeds. Diogenes then plucked a chicken and said: “Behold — Plato’s ‘man.’” We can keep adding qualifiers to this definition until we can no longer find a description that fits only humans and no other animal. Humans and animals are certainly different — of course we can find the most fitting description of humans from many perspectives. But isn’t that a bit too subjective?

Although this book refutes various claims of difference, the author does not deny that humans are special. In some respects, we are clearly unique. But we have yet to find that distinguishing point — at least, no consensus has been reached. If we want to find the essential difference between humans and animals, we must first discard the presupposition that “humans are special.”

Closing
#

First, a complaint about the Chinese translation — it screams machine translation. For example: “人们认为动物善于学习行为的普遍后果，但无法记住任何特定的联系” (“People believe animals are good at learning the general consequences of behavior, but cannot remember any specific connections”). It’s very hard to understand this sentence using direct Chinese thinking — it reads exactly like a machine-translated sentence. But if you think in English, it’s instantly clear: the sentence means “People believe animals are good at learning the consequences of behaviors but do not know the connection between the behavior and the consequence” (the author is refuting this statement).

Are We Smart Enough to Know How Smart Animals Are? has a strong academic atmosphere. It uses a wealth of reliable experiments and observations to explain the essence of animal behavior. Reading this book feels a bit like reading a paper — logically rigorous and cautiously worded in its claims. Primate studies, as the frontier of animal behavior research, hold great significance for studying human behavior — though some other animals’ behaviors are also useful.

The book contains many ideas that spark sudden flashes of insight: Clever Hans, the impossibility of equal testing environments for human infants and apes, the homology of all vertebrate brains, chimpanzees’ astonishing memory and logical reasoning abilities, chimpanzee power struggles, and more. Frans de Waal’s other book Chimpanzee Politics has already been added to my reading list…

Book Notes — Chimpanzee Politics: Power and Sex among Apes

Mon, 12 Aug 2024 00:00:00 +0000

Preface
#

Frans de Waal’s seminal work Chimpanzee Politics was published in 1982 — his first book and also recommended reading for incoming members of the U.S. Congress. Another work of his I read previously, Are We Smart Enough to Know How Smart Animals Are?, was from 2016 — such a vast timespan between them. Are We Smart Enough introduced many animal behaviors, including those of humanity’s numerous close relatives, while Chimpanzee Politics focuses solely on our very closest relative — the chimpanzee. It observes a chimpanzee colony in a zoo and analyzes the structure, evolution, and behaviors of chimpanzee social power and politics.

If you see chimpanzees at the zoo mating brazenly in broad daylight without any inhibitions, or screaming and attacking one another — seemingly devoid of moral restraint, showing no trace of civilization — then the English title of Are We Smart Enough serves as a perfect retort: “Are We Smart Enough to Know How Smart Animals Are?”

Power and Alliances
#

It’s commonly assumed that in animal social structures, the strongest male becomes the leader. This does broadly align with chimpanzee social structure. But it’s far from that simple — physical strength is not the sole factor determining dominance relationships. Alliances are the crucial factor, perhaps the most important factor. The book spends extensive passages discussing “triangular relationships.” Here, I need to introduce the book’s three main chimpanzee protagonists:

Yeroen (the elder) — Luit (the middle) — Nikkie (the young)

These three male chimpanzees form a power center — the power core of this chimpanzee colony — and their political struggles play out on this political stage. All three have, at different times, been the colony’s alpha. Initially, the capable and broadly respected Yeroen was alpha. Then Luit took over. Finally, Nikkie established a puppet-style rule. They built a hierarchical organization and competed within it for dominance over the rest of the group.

First: a male with superior fighting ability cannot simply usurp the group’s leadership. Power collapses not when a challenger defeats the current ruler in combat, but when the ruler can no longer protect other members of the society. During Luit’s bid for power, Luit and his ally Nikkie constantly attacked other group members, and when the Luit-Nikkie alliance was present together, Yeroen could not offer protection to others.

The Luit-Nikkie alliance played a decisive role in toppling the Yeroen dynasty. But Yeroen’s fall from power also created new alliance opportunities — just like human politicians, chimpanzees seize such opportunities too. Yeroen found the key player in the current “triangular relationship”: Nikkie.

Before Yeroen’s fall, Nikkie was Luit’s ally. Afterward, Nikkie became Yeroen’s ally. Why would the seasoned Yeroen support Nikkie after losing power?

For Nikkie: he went from number two to number one. He was the “person” most eager for Yeroen’s support.
For Yeroen: an alliance with Nikkie secured his position as number two in the group, and Nikkie — relative to Yeroen — needed his support more. Nikkie couldn’t openly oppose Yeroen, because if he did, Nikkie’s own position would become unstable. Yeroen gained more freedom of action and traded it for more mating opportunities with females.
As for Luit: he dropped from the top of the power rankings to number three.

The Yeroen-Nikkie alliance, though tight, featured a very cunning Yeroen. Although Yeroen’s relationship with Luit was terrible, Yeroen would still proactively approach Luit — and Nikkie would invariably intervene, without exception. Why did Yeroen seek contact with Luit? Yeroen approached Luit precisely to put on a show for Nikkie. For Nikkie, Yeroen’s behavior served as a constant reminder that Nikkie’s position depended entirely on Yeroen’s choices. The young Nikkie lacked strong grassroots support from the group. The seasoned, cunning Yeroen held Nikkie in the palm of his hand — Nikkie’s ruling foundation did not rest under his own feet.

When one chimpanzee grooms another’s fur, this is not merely a simple biological act — it’s a reflection of the two chimpanzees’ social relationship, signifying that their bond is sufficiently strong, or that one seeks a favor from the other. A classic scenario in the triangular relationship: Nikkie (center) grooms his ally Yeroen (left), while Luit (right) sits alone at a short distance.

Males and Females in Power
#

Although males are generally stronger than females, male chimpanzees do not use their full strength when attacking females. Males only bite and tear at each other when facing another male.

Social mammal groups are typically composed of many females and a few males. Females also play an important role in power struggles.

Female chimpanzees tend to avoid competition because they need a safer, more stable environment to raise offspring. Power transitions in the group do not happen instantaneously — when Luit replaced Yeroen, the process took over two months. During those two months, the two chimpanzees repeatedly fought and reconciled. Female chimpanzees played a vital mediating role in this process. Females would proactively embrace both of them, breaking the tension during confrontations and working hard to push them toward reconciliation.

Male leadership arises from strength, alliances, and support levels. Females also have a leader, but female leadership is determined by character and age. Females almost never need to fight each other; the probability of conflict between females is extremely low, and their hierarchical order can persist for many years.

Social psychologists, through alliance-game testing, have found that males take more proactive action, while females place more emphasis on the atmosphere of the game. In competitive activities, men are all about achieving strategic objectives — they prefer to seize the “big” events. Women are more interested in individual connections, forming alliances with those they like, and they focus on the immediate rather than distant political goals. Of course, these are statistical tendencies — exceptions always exist.

Power and Sex
#

Avoiding incest is a moral or legal constraint in human society, often considered part of human culture. If mating were purposeless, would group-living chimpanzees have incest problems? In reality, such problems are extremely rare. Chimpanzees actively avoid incest. Mothers know who their sons are, and when a son reaches adulthood, chimpanzee mothers absolutely will not tolerate incestuous behavior. Young chimpanzees may not know who their fathers are, but they strongly resist mating with males roughly their father’s age. Biologists believe incest avoidance is a natural law deeply embedded in culture.

Power and sex are certainly linked. Chimpanzee alphas typically enjoy extremely high mating privileges — until overthrown by a rebel. But these mating privileges occur during ordinary times; female chimpanzees will secretly mate with males they treated coldly during the day — at night, or in places the alpha can’t see, like in the tall grass. How similar this is to human society needs no elaboration.

Jealousy produces more offspring. Chimpanzee social structure includes multiple females and males. More jealous males will do everything to prevent other males from contacting females, giving themselves more opportunities to sire offspring — and those offspring, in turn, will also be more jealous. Females, however, are entirely different: no matter whom she mates with, her number of offspring is fixed, and the offspring are always hers. So jealousy among females is not pronounced. But in pair-bonding species, things look completely different — in pair-bonding species, females also engage in sexual competition. In such cases, females are more inclined to maintain long-term relationships with males. In modern human society, men care more about whether their female partner has had sex with another man; women care more about whether their partner has fallen in love with another woman. At its essence, even the cornerstone of human society — the family — is merely a unit of sex and reproduction.

Closing
#

There’s actually a lot more interesting material I haven’t gotten to — too lazy to expand further. Some perspectives I personally really like:

“Humans are engaged in continuous office competition while simultaneously uniting against a common enemy.”
“Hierarchical order is a cohesive factor that imposes limits on competition and conflict.”
“The roots of politics are far older than humanity.”

Universal Safety Disclaimer
#

Large portions of this article are drawn from the book Chimpanzee Politics and do not represent my personal views.

Book Notes — Homo Deus: A Brief History of Tomorrow

Mon, 12 Aug 2024 00:00:00 +0000

Preface
#

Homo Deus: A Brief History of Tomorrow is one of the trilogy by Israeli historian Yuval Noah Harari. The trilogy consists of Sapiens: A Brief History of Humankind, Homo Deus: A Brief History of Tomorrow, and 21 Lessons for the 21st Century. The most famous, of course, is Sapiens — an extraordinarily sweeping book about the history of human civilization that can absolutely reshape your view of history. Last year (2022), I stubbornly gnawed through the English original of Sapiens page by page — quite an achievement. Because I loved Sapiens so much, Homo Deus, the sequel from this giant of a thinker, naturally became this year’s most important “extracurricular reading.”

Sapiens tells the story of human history — from Homo sapiens standing upright to launching rockets to explore the stars: how did we get here? Homo Deus discusses the critical issues currently facing human civilization, and where we are headed.

This copy of Homo Deus was hard to come by. In the end, I bought a second-hand Chinese edition from JD — it came from the library of Xingtan Liang Qiuju Middle School 😄. When I opened to the first page, a cheeky middle schooler had left a line of English. Let’s start with that:

When facing the ultimate questions of this chaotic world, we need Chinese readers to contribute their wisdom.

The New Agenda
#

Famine
#

Open almost any history book, and you’ll read about the horrors of famine and the insane behavior of people pushed to starvation. There’s no need to bring up famines in other countries — the most noteworthy case is right here in China. From the earliest written records all the way to the 20th century, China suffered the ravages of famine for thousands of years. We’ve always been an agricultural nation; nearly everyone had to work the land to feed themselves and their families. If crops failed — due to natural disasters (too much or too little rain, locust plagues, etc.) or human interference (bandits, oppressive taxes, irregular planting) — some people would face food shortages. Most modern people have no idea what it feels like to go without food for days on end. I’ve been hungry for stretches myself, and I know that prolonged hunger is a misery the average person can’t imagine — but even I was never at risk of starving to death. Yet our ancestors, facing the prospect of actually starving to death, what kind of despair must they have felt? They had no solution but to pray to the gods for favorable weather and a bountiful harvest the following year.

There’s a line from House of Cards that really stuck with me: “Twenty years ago, I couldn’t buy sugar in China. Now I can buy it anywhere.” Crude as it sounds, it reflects a reality: the Chinese people have escaped poverty. For the first time in Chinese history, we are no longer tormented by famine. We created this economic miracle — something worth recording! Similarly, human civilization as a whole has recently solved the problem of hunger. Food shortages in particular regions are almost always caused by political factors, and internationally, there are ample surplus resources for emergency response to shortages. Food scarcity is no longer a human agenda item.

On the contrary, humanity is no longer concerned with food shortages but is starting to worry about food surpluses. Health problems caused by obesity and malnutrition far outnumber those caused by starvation. Many people mindlessly chew through bread, rice, and loads of carbohydrates without getting enough protein and vitamins. The rich eat lettuce salads; poor Westerners eat cake, burgers, and pizza; and I eat fried dough sticks, steamed buns, rice, and noodles — my weight keeps climbing every day, and my health problems multiply year by year…

Bacteria and Viruses
#

The Black Death: In the 1330s, the Black Death — the bacterium Yersinia pestis — caused 70 to 200 million deaths worldwide, with a mortality rate of roughly 50%.
The Spanish Flu: 1918. Infected 500 million people; 50 to 100 million died. Mortality rate around 15%.
Smallpox: 1967 — 15 million infected, 2 million deaths. Mortality rate about 15%. Following global smallpox vaccination, the smallpox virus was eradicated by humanity in 1979.
AIDS: Broke out in the 1980s. Over 30 million deaths. Destroys the immune system. Current medications are effective but cannot provide a perfect cure. Infection rate: 0.9%. Mortality rate: 1.28 per 100,000.
SARS: 2003. 8,000 infected, over 700 deaths.
Avian Flu: Fewer than 1,000 deaths.
H1N1 Swine Flu: 2009. 700 million to 1.4 billion infected. Approximately 150,000 to 600,000 deaths. Infection rate: 20%. Mortality rate: ~0.02%.
Ebola: Multiple outbreaks in Africa. Mortality rate above 50%.

Of the above, only the Black Death is bacterial; all the rest are viral. The Black Death is too ancient; though bacterial, due to the primitive state of medical care at the time, people had no idea what was happening, leading to massive casualties and an extraordinarily high fatality rate. Smallpox is humanity’s greatest success story in the war against viruses — through modern medicine and vaccines, we outright eliminated the smallpox virus. As you can see, humanity has developed a silver bullet for bacteria — antibiotics. Bacterial epidemics are essentially gone. But for viral influenzas, they keep emerging in an endless cycle: as one subsides, another rises. There’s no great solution; modern medicine still has room to improve against viral epidemics. Major viral pandemics still strike every few years, and seasonal flu never stops accompanying us.

Most of these influenzas are weathered by the human immune system alone — modern medicine only plays a supporting role (basically, bringing down fevers). Especially for human infants: aside from getting all manner of vaccines right after birth, every other “cold” has to be tough out by their own immune systems, with very few effective medications available. Kindergarten is less a place of learning and more a trial ground for human influenza and immune resistance.

COVID-19: Homo Deus was published in 2015, before COVID-19 happened. The author’s view on epidemics was: “Doctors can quickly get up to speed and rapidly discover treatments — humanity has probably already conquered epidemics.” I wonder what Yuval Noah Harari makes of COVID-19.

Regarding COVID-19, there’s simply too much I want to say. So many grievances that they defy coherent complaint. In a single sentence: “On the matter of COVID-19, humanity was utterly shattered and exposed in all its ugliness.”

War
#

Skipped.

Humanism
#

The concept of humanism emerged during the Renaissance, championing human rights and individual value in opposition to the religious theocracy of the time. It reached China roughly in the late Qing dynasty. Humanism has had a profound impact on modern society: people increasingly emphasize the concepts of the individual or the collective, rather than top-down religious dogma and the divine right of kings.

Humanism advocates human rights against divine right, and individual freedom against personal dependency. What humanism worships is human nature — the human being itself.

People are always contemplating the meaning of life. Humanism holds that humanity itself is the source of meaning: “I am the meaning.” It also holds that free will is the highest authority. Humanism proposes a new life principle: “If I feel it’s good, it’s good; if I feel it’s bad, it’s bad.” For example: if a woman has an affair, in pre-humanist society, she would face punishment from religion and social norms — the censure of priests and elders. In modern society, she need only heed her own true feelings; the best approach is to ask her own heart what it thinks.

For society as a whole: what everyone believes is good is “good”; what everyone believes is bad is “bad.” Take theft, for example. For the victim, it’s certainly bad. For everyone else, it’s also bad — because others don’t want to be stolen from either, including thieves themselves. Thus, theft is bad, and people can even write it into a mutually binding document. By the same logic, if a certain behavior feels bad to no one at all, then it’s not wrong. This naturally leads to the question of homosexuality: two people of the same sex feel that this is good, and it affects no one else — therefore, it’s not wrong. So humanism supports homosexuality and opposes religion.

Humanism can perfectly address these two types of extreme questions. But for events that are good for some and bad for others — like the trolley problem — it’s much harder to answer. In ancient societies, Confucianism advocated that women remain faithful to one husband unto death, even erecting chastity archways. In modern society, as long as one can find happy days, people don’t want to stay bound in misery. But what if divorce leads to happiness for one side and utter misery for the other? Add the emotional harm to the children, and the whole situation becomes very hard to measure: whose happiness matters more? Humanism will only tell you: “Follow your own heart."~

As humanism gained broader acceptance, it evolved into three major branches:

Liberal Humanism: The “orthodox” liberal humanism, also known as liberalism. The individual enjoys freedom; individual choice is respected. If it feels right to each person, it’s right. The classic example is liberalism’s belief that the ballot box represents individual will. But this requires one precondition: before voting, everyone must be “one of us.” For instance, the American North and South in 1861, or Israel and Palestine today — neither could possibly resolve their issues by having everyone vote together.
Socialist Humanism: Socialist humanism doesn’t focus on individual feelings, viewing them as a bourgeois trap. What “I” feel in the present moment is merely a reflection of my environment, determined by my class. Liberalism believes voters can make the best choice; socialist humanism believes the organization can make the best choice. The individual must obey the organization’s decisions, not personal feelings.
Evolutionary Humanism: Evolutionary humanism derives from Darwin’s theory of evolution. It holds that conflict is a form of evolution — eliminating the weak, survival of the fittest. Superior people deserve to survive; this is the law of human evolution. Evolutionary humanism was once all the rage, giving rise to many ideas such as eugenics, racism, and fascism.

From 1914 to 1989, the three humanisms waged a war of faith. Liberalism and socialism joined forces to defeat Nazism in World War II. Then liberal nations and the Soviet Union each rallied allies into the Cold War. In the early Cold War, socialism consistently held the upper hand (the documentary The Vietnam War is highly recommended here) — students at UC Berkeley even kept Chairman Mao’s Little Red Book by their bedsides. Then, everything changed. The Soviet Union collapsed. Many countries shifted their beliefs; we too introduced market capitalism. People preferred supermarkets (or Taobao) and money-making companies over a system that allocated food and clothing. Liberalism won a sweeping victory in this war of faith — they even evolved further, adopting ideas and institutions from their rivals to provide better education, healthcare, and social security than before. But liberalism’s core ideology remained unchanged.

Dataism
#

Dataism holds the following three views:

Organisms are algorithms.
Intelligence can exist without consciousness.
Highly intelligent algorithms know me better than I know myself.

Organisms Are Algorithms
#

“Organisms are algorithms” — I couldn’t accept this notion when I first encountered it either. How could organisms be algorithms? Doesn’t human experience matter? Is human consciousness worthless?

Looking at capitalism and Soviet-style communism from the perspective of data processing, they are no longer ideological opposites but rather different data algorithms. Capitalism employs a distributed algorithm; Soviet-style communism employs a centralized algorithm. Capitalism allows connections between consumers and producers, permits individuals to freely exchange information and make independent decisions — the pricing and output of goods are determined by the free market. Soviet-style communism, on the other hand, severed the link between producers and consumers: the government collected consumption data and issued production directives to producers. The government took all of the workers’ productive surplus, then determined what each individual needed, then re-distributed accordingly. Tax rates work the same way — high tax rates essentially concentrate more resources together, with the government as a single processor deciding how resources are allocated and utilized.

A single processor can’t possibly make the right decisions forever. No one person can handle such enormous amounts of data — even today’s high-speed computers can’t process it all.

From the perspective of Dataism, capitalism won the Cold War because its distributed algorithm was better suited to that era than Soviet-style communism’s centralized algorithm: the better data algorithm prevailed. When we chose to embrace the market economy and abandon Soviet-style communism, it was equivalent to decentralizing processing power to every individual, no longer using the single-processor model. That’s why Socialism with Chinese Characteristics survived the Cold War, while the Soviet single-processor data model failed utterly. Currently, only a very few authoritarian states still use this single-processor model — and after all these years, we’ve seen no productivity advances from them. This is also a real-world reflection of “organisms are algorithms.”

I’m merely using the Dataist lens to view economic models here — no intention of judging which model is better or worse. Beyond fitting economic models so neatly, Dataism can also be applied to view problems in many other domains.

Intelligence Can Exist Without Consciousness
#

First, we need to be clear: what is consciousness? Someone might say, “Consciousness is the self,” or “Consciousness is the voice inside.” These don’t answer the question scientifically. Note: science deals with objective facts; subjective matters fall outside science’s domain — they belong to theology. We cannot explain the subjective using the subjective. In truth, humanity still hasn’t figured out what consciousness is.

If every person is an algorithm, then there’s really no concept of “autonomous consciousness.” We can regard what we hear, smell, and see as “input data.” After computation by our biological organism, a response is produced and an action taken — that’s the “output data.” The human body itself is more like a CPU — perhaps one that can self-regulate, but even the regulation itself requires data input, like learning knowledge or exercising. So what role does “self-consciousness” play in this process? I can clearly make choices about something — if I choose differently, a different outcome results. I must be consciously aware… right? This question may not be so easy to answer. If, hypothetically, there were no subjective consciousness — not brain death, but “I can’t feel my self” — would “I” still make different choices?

From a biological perspective, consciousness is nothing more than countless electrical currents in the brain’s neural network. When “I” make a different choice, it may simply be that some nerve ending fired an extra tiny electrical pulse. “Self-consciousness” played no role whatsoever in this process. Without “me,” it seems my body could still make different choices, as long as the “algorithm” stored in my body still exists. If “self-consciousness” exists, it’s more like a belief rather than an objective fact — like believing in God. The most cutting-edge biological science suggests that consciousness is merely a byproduct of an individual organism’s algorithms — it could even be viewed as a kind of mental pollution.

Then we arrive at another question: is artificial intelligence (AI) conscious? If it’s not conscious, can we treat it as an intelligent being? The best method humans currently have for testing whether AI has consciousness is the Turing Test. The Turing Test’s logic is simple: as long as a normal human can’t tell whether the AI is human or not, it passes. In other words, once AI becomes smart enough, we humans have no choice but to consider it “conscious.”

Algorithms Know Me Better Than I Know Myself
#

Humanism calls on us to listen to our inner authentic voice. But if the self doesn’t even exist, what is there to listen to? Dataism calls on us to “listen to the algorithm’s advice” — the algorithm knows me better than I know myself. For example: when a woman is on a blind date and meets two men who both seem suitable, without algorithmic assistance, she would follow her inner voice and choose the one who “feels” more right. Now imagine an algorithm tells her: “I know you very well. I know you’re attracted to Man A; you’ll choose him. But he will ultimately break your heart and leave you. Man B is the one for you — and if you choose B, you’ll fall in love just as quickly. He will give you lasting happiness. This is a choice you won’t regret.” From any angle, shouldn’t she listen to the algorithm’s advice rather than that fleeting feeling of the moment?

In its early R&D phase, algorithms are built by engineers continuously piling up code. At this stage, people still have a decent grasp of what the algorithm is “thinking.” But algorithms can self-learn and self-update. Their learning capacity is utterly beyond human comparison. They will gradually carve out their own path, until humans can no longer keep up.

Closing Thoughts
#

Homo Deus is, as ever, packed with substance — novel, robust ideas, all-encompassing. A highly recommended work. While reading, I often paused to reflect: does what he’s saying match reality? Is it correct? Many times, I felt shocked. I used to never underline when reading books, but I did so with this one. When I finished, I found the book covered in my highlights.

This monumental work is so dense with content that this article can’t possibly cover everything. This piece is relatively one-sided — I’ve mostly only discussed productivity-related viewpoints. There’s actually a great deal of other fascinating material, such as the book’s perspective on “happiness”: “Would you rather be an unhappy but wealthy Singaporean, or a happy but poor Costa Rican?” I don’t know how I’d answer. But if the author rephrased the question to me as: “Would you rather eat more hot pot, or eat vegetables and whole grains every day, maintaining a nutritionally balanced, healthy body?” — then I’d definitely answer: hot pot.

Book Notes — Romance of the Three Kingdoms

Mon, 12 Aug 2024 00:00:00 +0000

Preface
#

Mention Romance of the Three Kingdoms, and it seems almost everyone can name a few characters or plot points. But have you actually read the original?

I’ve always been a fan of Three Kingdoms-themed games — titles like Bàwáng Dàlù (The Overlord’s Continent) and Total War: Three Kingdoms are among my favorites. I love the feeling of collecting famous generals and rampaging across the battlefield. But thinking back, I realized I’d never actually read Romance of the Three Kingdoms in its entirety. Some of those generic officers in Total War — I had no idea who they were. And when I thought about it, I couldn’t come up with a single novel that could stand toe-to-toe with Romance, so I decided to give the original a try. Once I started, I couldn’t stop…

Romance of the Three Kingdoms is written in the vernacular Chinese of the ancient period, which differs somewhat from modern vernacular Mandarin. For example, “暗赍金帛，结交中涓封谞” means secretly bringing gold and silk to befriend the eunuch Feng Xu (赍, pronounced jī, means “to bring” — a common exam term; 中涓 was a close-attendant official title, later used to refer to eunuchs in general). At first, it was admittedly hard going, but after a while it became quite smooth. When I didn’t understand something, I’d just check the annotations or underline it (once again, thank you, e-books). Also, a reading tip: skip the preface. I recommend mentally filtering out keywords like “peasant uprising”, “dialectical”, “feudal”…

Many people confuse Romance of the Three Kingdoms with Records of the Three Kingdoms. Let me emphasize this: Romance of the Three Kingdoms is a novel; Records of the Three Kingdoms is an official history. Some might counter, “But the Records was privately compiled,” or “There is no single truth in history.” You can believe there’s no absolute truth in history, but if you carry that attitude into historical scholarship, then there’s no point studying history at all. Not every statement in an official history is precise — some contain ambiguous or even contradictory accounts — but that only affects its reference value, not its status as an official history. The Records of the Three Kingdoms is one of the Twenty-Four Histories, an undisputed official history, beyond all doubt. Romance of the Three Kingdoms is a novel written with deep reference to the Records, upon which artistic embellishments were layered.

Unless I explicitly mention the Records, this piece discusses the novel alone. Although I dipped into the Records (and discovered that Total War draws primarily from the Records 👍), I found it too hardcore and decided to give up. In any case, the novel’s characters and plotlines all involve artistic license and differ from history — readers, please keep the distinction in mind.

Red Cliffs
#

There are many plotlines in Three Kingdoms worth discussing, but given space constraints (or, honestly, I just don’t feel like writing more), I’ll focus on Red Cliffs.

The Battle of Red Cliffs is undoubtedly the crown jewel of the novel. All the great names make their entrance, stratagems fly thick and fast, and the intellectual duels between Zhou Yu and Kongming (Zhuge Liang) elevate the battle to the pinnacle of wit. I’ve always been fond of Zhou Yu — brimming with talent, dashing and heroic, brave and resourceful, commanding armies with brilliance, achieving greatness young (a winner in life), with the ability of a king’s right-hand minister. But to highlight Zhuge Liang’s genius, the novel deliberately places Zhou Yu’s talents a notch below Kongming at every turn, making him Red Cliffs’ absolute foil to set off Zhuge Liang. As I watched the TV series and read the novel, I increasingly felt that the early-period Zhuge Liang was simply a “monster” — “utterly inhuman.”

Red Cliffs features an all-star cast. All three of Liu Bei’s top strategists were involved: Zhuge Liang, Pang Tong (then serving Wu), and Xu Shu (then in Cao Cao’s camp) each played critical roles in the battle’s schemes. The warriors basically just cleaned up — only Liu Bei and Guan Yu visited Wu’s naval camp, and Zhao Yun once rescued the strategist. Wu itself was the protagonist, naturally — Zhou Yu, Lu Su, Huang Gai, Gan Ning, Kan Ze all had major parts, and the rest of the Wu officers were united in purpose, not a single one dragging their feet. On Cao Cao’s side: the Chancellor himself, advisers Cheng Yu and Xun You (Xun Yu and Jia Xu didn’t come; Guo Jia had died young), officers Mao Jie and Yu Jin, famous generals like Zhang Liao and Xu Chu essentially making cameo appearances, plus the tragic patsies Cai Mao, Zhang Yun, Cai Zhong, and Cai He, and the clown Jiang Gan… I suspect many people have never read Red Cliffs carefully, or perhaps never read the original at all. I specifically drew a flowchart:

Two favorite passages:

“A gust of wind blew, lifting a corner of the banner to brush across Zhou Yu’s face. Yu suddenly recalled something weighing on his heart, let out a great cry, fell backwards, and vomited blood.” This brief passage is gripping, vivid as a film scene, and underscores the importance of the southeast wind — not a single word wasted.

Yu said: “‘Man’s fate shifts between morning and evening’; how can one guarantee one’s safety?” Kongming smiled and replied: “‘The heavens hold storms none can foresee’; how can man predict them?” Yu turned pale upon hearing this and feigned moans of pain… Kongming smiled: “I have a prescription that will settle the Commander’s distress.” The entire exchange never once mentions the east wind, yet Yu and Liang have already dueled several rounds over it. Truly brilliant~

Embellishments
#

Adding one’s own interpretations or plot elements on top of the original — I call these “embellishments.” The original plot is already extraordinarily compelling. Even where modern readers might find things hard to understand, if you immerse yourself in the mindset of ancient (Eastern Han!) people, there are virtually no logical gaps. This is one reason why Romance of the Three Kingdoms is held in such high regard. That’s why many people still prefer the old TV adaptation that respects the original (with minimal changes) over the new adaptation full of embellishments. Some embellishments — no one even knows who started them — conspiracy theorists abound, and many fabricated plotlines have become widely accepted as fact, which is truly a shame.

Kongming letting Cao Cao escape. At Huarong Trail, where Guan Yu spares Cao Cao out of a sense of honor, Kongming deliberately sent Guan Yu knowing Cao Cao would be released. This has spawned countless interpretations. But in the original, Cao Cao escapes simply because Kongming, observing the stars at night, concluded that Cao Cao was not fated to die that night. Don’t dismiss this as childish — the novel treats “star-reading” as a very real mystical phenomenon. “Read the stars and release Cao Cao” — this reason is entirely sufficient in the novel’s own terms. As for “they feared Wei’s retaliation so they let Cao Cao go” — pure later embellishment. Romance never once features a plot where someone refrains from killing out of fear of retaliation. The same goes for Guan Yu’s death.

Guan Yu’s death. In the original, both Wei and Wu wanted Guan Yu dead — such was the era, and in the novel, Guan Yu was a godlike figure. You couldn’t take Jing Province without killing him; both sides went all out. It’s true that later, when Liu Bei raised a great army for revenge, both Wu and Wei tried to pass the blame — but that’s all after Guan Yu’s death. Also, many later commentators believe Guan Yu should have defended Jing Province rather than attacking. Here I must clear General Guan’s name: attacking Fancheng was Zhuge Liang’s order. Guan Yu simply failed to take it.

Pang Tong’s death. The original says Kongming, observing the stars at night, saw a general’s star falling and sent a letter warning Liu Bei to be cautious. But Pang Shiyuan (Pang Tong) suspected Kongming was just afraid of him stealing glory and urging Liu Bei to advance slowly, so he in turn pressed Liu Bei to speed up the campaign — and ultimately died at Fallen Phoenix Slope. The new Three Kingdoms TV series embellished this: Liu Bei couldn’t bear to seize Yi Province, he knew there was an ambush but still entered Fallen Phoenix Slope, sacrificing himself to give Liu Bei a pretext to break with Liu Zhang… (this embellishment honestly disgusts me). Liu Bei and Liu Zhang’s conflict actually escalated gradually — Liu Zhang’s subordinates were already fighting Liu Bei, but the final break came when Liu Zhang discovered Zhang Song’s letter of surrender and realized Liu Bei’s wolfish treachery. While we’re here, let’s discuss a frequently debated detail: did Liu Bei give Pang Tong the Dílú horse? (The Dílú was said to bring misfortune to its rider; Xu Shu once advised Liu Bei to gift it to an enemy to avert the curse, then ride it himself — but immediately said he was merely testing Liu Bei’s character.) I’ve seen many comments assuming Liu Bei gave Pang Tong the Dílú, but reading the original carefully, it’s actually quite ambiguous. Liu Bei gave Pang Tong a white horse, but it’s never specified as the Dílú. In fact, after leaping across Tan Stream, the Dílú basically vanishes from the story. If it truly brought misfortune, Liu Bei had given it to Liu Biao (who returned it upon learning of the curse) — and Liu Biao died anyway. If it didn’t truly bring misfortune, that’s also plausible. Romance doesn’t treat all mystical elements as absolute truth: believing in them can be called respecting the spirits; disbelieving can be called being an extraordinary man or hero. Liu Bei and the Dílú lean more toward the latter, because Xu Shu was really just testing Liu Bei’s benevolence: “A man’s life and death are determined by fate — how could a horse be the cause?” If the horse truly brought misfortune, Xu Shu wouldn’t have said “I was testing you.” So personally, I believe what Liu Bei gave was not the Dílú, but simply one of his ordinary white horses. For a lord to gift his own horse was an immense honor in ancient times — this was simply meant to show Liu Bei’s genuine affection for Pang Tong. Later generations just preferred the Dílú storyline and embellished accordingly.

Diaochan’s righteousness. Only very rarely do embellishments improve things. In the original, the eighteen lords’ coalition was utterly helpless against the Western Liang army. After entering Luoyang, they all went their separate ways — while the Emperor remained in Dong Zhuo’s clutches… And then, contrast this with Diaochan, a mere woman: solely to repay Minister Wang Yun for raising her (Diaochan was his adopted daughter), she offered her body and successfully drove a wedge between Dong Zhuo and Lü Bu. After this, the novel mentions Diaochan very little (she simply follows Lü Bu, with no further plot involvement). The Three Kingdoms TV adaptation’s treatment of Diaochan after her success is truly brilliant. The old Three Kingdoms series adds an epilogue for her: to a hauntingly beautiful melody, Diaochan retreats into obscurity after her great deed, never to be heard from again. The fate of a nation rested on a frail woman — starkly contrasting with the warlords’ failure against Dong Zhuo and their secret scheming against each other. This segment is exquisite. Diaochan is a true hero! Compare this to the new Three Kingdoms’ treatment of Diaochan: pure schlock, fabricating a romance between Lü Bu and Diaochan — utterly an embellishment, disrespecting the original and even disrespecting Diaochan.

Mysticism
#

It’s a novel, after all — many plot points are dramatized additions (the same goes for Water Margin and others). A bit of artistic license for reading pleasure is “the finishing touch on a dragon painting,” not “drawing legs on a snake.” Personally, I prefer to read Romance of the Three Kingdoms as a fantasy novel rather than a historical one.

The Yellow Turban Rebellion. The Yellow Turban Rebellion was less a peasant uprising than a religious war. At first, seeing Zhang Jiao cure people with talisman water, I assumed the author was portraying the Yellow Turbans as uncivilized charlatans. Then I discovered that Yu Ji also cured people with talisman water — and Yu Ji is clearly a positive character. Sun Ce disbelieved and ended up being mystically killed by the Little Conqueror. So talisman-water healing is a real thing in the author’s universe. The three Zhang brothers genuinely possess supernatural abilities, and the Yellow Turban army is basically a religious sect. I eventually accepted the talisman-water premise.

“His ears hung down to his shoulders, his hands reached past his knees, and his eyes could see his own ears.” Hands past the knees, fine — but eyes that can see your own ears? That’s not an eye problem, that’s an ear problem. The man was probably an elephant…

Killing one’s wife for food. While fleeing and seeking sustenance, Liu Bei encounters a hunter. Having found no game, the hunter kills his wife and serves her as food. Liu Bei only realizes the previous night’s meal was the man’s wife: “overcome with sorrow, he shed tears and mounted his horse.” When Cao Cao hears of the “kill-wife-for-food” incident, “Cao ordered Sun Qian to reward him with a hundred taels of gold.” Even someone like me, with fairly open views, was utterly shocked reading this. It’s astonishing how different ancient values were from ours, and lamentable how low women’s status was — mere objects…

Star-reading. In ancient times, star-reading was an official government post. In Romance, it’s a skill possessed by high-level strategists. Pang Tong lacks star-reading ability; Zhuge Liang and Sima Yi possess it.

Rǎng (禳). Actively attempting to alter fate. When Liu Bei’s Dílú horse threatened misfortune, the method Xu Shu described to dispel the calamity was called a rǎng ritual. Zhuge Liang used the qí-rǎng ritual to pray to the Northern Dipper, seeking to extend his life by one jì (twelve years).

Flaws
#

Some important characters are described too sketchily. “By the time Song arrived, Zhang Jiao was already dead” — just a handful of words dismiss the death of the leader who ignited the earth-shaking Yellow Turban Rebellion. I always find this hard to accept; the author doesn’t even tell us how Zhang Jiao died. (If you Baidu it, you’ll just get middle-school history memorization paragraphs about why the Yellow Turban uprising failed…)

Some plotlines are repetitive. The famous “Borrowing Arrows with Straw Boats” actually appeared earlier. Sun Jian, while attacking Huang Zu, had a similar arrow-borrowing episode: “Jian plucked the arrows embedded in his boats, amounting to over a hundred thousand.” Red Cliffs and Yiling share similarities too — “southeast wind,” “boats loaded with thatch,” and “fire attack” are all keywords of Yiling as well. The Girdle Edict in the early chapters is a compelling storyline, and later there’s a parallel with Wei Emperor Cao Fang’s blood-written edict.

After Zhuge Liang’s death, the later plot isn’t very engaging. By then, almost everyone I knew was dead. There’s Jiang Wei and Deng Ai to follow, perhaps, but the plotlines are formulaic and dull. The new characters are numerous but lack distinctive portrayals — you basically can’t remember them. Later battle scenes all follow the same template: feign defeat, lure the enemy deep, a cannon blast, then charge.

Character Biographies
#

A sharp-tongued review of several characters, with brief introductions, summaries, and key deeds. Though it doesn’t quite align with the novel’s spirit, people always love debating martial prowess and intelligence scores. Having read the entire novel, I’ll try to discuss the numbers here too.

For one-on-one combat ratings, it’s not about who defeated whom — Romance features many draws, or fights broken off after twenty or thirty bouts for various reasons. I measure by number of bouts exchanged. In Romance, 100 bouts is generally the upper limit; fighters may rest and resume for another 100, as with Ma Chao and Xu Chu.

Wei Side
#

Cao Cao: Military strategist, statesman, man of letters. Extraordinarily fond of talent, shrewd himself, maxed out in both intelligence and ruling ability. The man who won the Central Plains battle royale. Welcoming Emperor Xian and establishing military farms (túntián) were both pivotal moves. There’s too much to say… Everyone knows “a crafty hero in turbulent times,” but few mention “an able minister in peaceful times.”

Xun Yu: Cao Cao’s key early strategist, intellect no less than Guo Jia. Loyal to the Han dynasty to the end. Killed by Cao Cao.

Xun You: Wei’s strategist in the humiliating Red Cliffs campaign. Has intelligence but a notch below Guo Jia and Xun Yu.

Guo Jia: Flawless. The number one grand-strategy adviser, relying on intellect rather than mysticism. Universally beloved. Died of illness while accompanying Cao Cao on the northern campaign. After Cao Cao’s defeat at Red Cliffs, he wept that Fengxiao (Guo Jia) was no longer with them — all others hung their heads in shame.

Cheng Yu: A strategist who appeared frequently in the early period. High intelligence — personally, I’d rate him roughly on par with Cao Cao: top-tier, but below Xun Yu and Guo Jia. At Red Cliffs, he saw through the southeast wind issue but was talked down by Cao Cao.

Jia Xu: Adviser to Li Jue, later joined Zhang Xiu, later surrendered to Cao Cao. An important mid-period Wei strategist.

Xu Chu: Wei’s top-tier solo combat god. Captured He Yi alive in one bout. Fought Ma Chao for 200 bouts — the bare-chested war god. During the Hanzhong campaign, drunk on grain-transport duty, he was slow to react and got stabbed in the shoulder by Zhang Fei. Limited appearances after that.

Dian Wei: Wei’s top-tier solo combat god. Master of twin halberds. Fought Xu Chu for two full shíchén (four hours). Felt like Cao Cao’s personal bodyguard. Killed during Zhang Xiu’s rebellion. Limited combat record.

Cao Ang: Cao Cao’s eldest son by Lady Liu. Killed during Zhang Xiu’s rebellion. Gave his horse to his father to ride, couldn’t escape himself. After the battle, Cao Cao wept only for Dian Wei, not for Cao Ang…

Cao Pi: Cao Cao’s eldest son by Lady Bian. One of the Three Caos. Proclaimed himself emperor immediately after Cao Cao’s death. Defeated at Hefei.

Cao Zhang: Cao Cao’s second son by Lady Bian. Has combat achievements — defeated Liu Feng in three bouts. A pure warrior archetype. “A real man should emulate great generals like Wei Qing and Huo Qubing, leading a hundred thousand troops across the desert, driving out the barbarians, building a legacy of achievement — who would want to be a scholar?”

Cao Zhi: Cao Cao’s third son by Lady Bian. One of the Three Caos. “Vain and flashy, lacking sincerity, addicted to wine and unrestrained.”

Cao Xiong: Cao Cao’s fourth son by Lady Bian. Killed in the power struggle when Cao Pi succeeded to the throne.

Cao Chong: Not mentioned in the novel.

Cao Ren: A commanding general, no solo combat record, but a master of city defense. (His troops) shot Zhou Yu at Nanjun; (his troops) shot Guan Yu at Fancheng. Died during Cao Pi’s reign.

Cao Hong: Often appears leading troops. Fought He Man for fifty bouts and killed him in single combat. Personally killed Yuan Tan. Rescued Cao Cao at a critical moment. Fought Ma Chao for fifty bouts — his blade technique grew disordered, his strength failing. With Cao Xiu, forced the Han Emperor to abdicate. No further appearances.

Xiahou Dun: Fierce and bold. Took an arrow to the eye and swallowed his own eyeball. Fought Gao Shun for fifty bouts — victory. During “Crossing Five Passes and Slaying Six Generals,” he challenged Guan Yu to a duel, interrupted by Zhang Liao. The Wei protagonist at Bowang Slope. Died of illness during Cao Pi’s reign.

Xiahou Yuan: Master of long-distance rapid strikes (couldn’t find the original quote). Many appearances leading troops — a general-type commander. Later killed by Huang Zhong at Mount Dingjun.

Zhang Liao: Leader of the Five Elite Generals. Formerly under Lü Bu; close friends with Guan Yu. First-rate at leading troops, decent at solo combat. While accompanying Cao Pi against Wu, shot in the waist and killed by Wu officer Ding Feng.

Zhang He: Of the Five Elite Generals. Framed by Guo Tu at Guandu; defected to Cao Cao. Defeated by Zhang Fei (at Zhang Fei’s tomb in Langzhong you can still see Zhang Fei’s inscription “Great Victory over Zhang He’s Forces”). Seems only able to trade a few dozen bouts with Zhang Fei — solo combat: average; commanding troops: first-rate. More appearances in the early period. Later, pursuing too deep, killed by Kongming’s massed crossbows at Jianmen Pass.

Xu Huang: Of the Five Elite Generals. Appears so often it’s impossible to recount everything. During Li Jue and Guo Si’s rebellion, served under Yang Feng, later defected to Cao Cao. Fought Xu Chu for fifty bouts — solo combat: decent. Also close friends with Guan Yu (during the Yan Liang-Wen Chou incident, Zhang Liao and Xu Huang fought poorly; Guan Yu stepped up and cut each down in one stroke — presumably this is when they became friends…). With Cao Ren, jointly defeated Guan Yu’s Jing Province army. Later, when Meng Da rebelled again, was shot in the forehead and died.

Yue Jin: Of the Five Elite Generals. Also appears very frequently, often leading troops. Fought Lü Bu’s officer Zang Ba for thirty bouts; fought Ling Tong for fifty bouts — solo combat: average. During the Hefei campaign against Sun Quan, while dueling Ling Tong, Cao Xiu shot Ling Tong off his horse; Gan Ning then shot Yue Jin in the face with a single arrow. Never appears again — unclear if he recovered.

Yu Jin: Of the Five Elite Generals. During Zhang Xiu’s rebellion, when people accused Yu Jin of defecting, he didn’t first clear his name but instead set up camp to resist the enemy — praised by Cao Cao. When Fancheng was besieged, he led reinforcements. Afraid Pang De would steal glory, he engaged in various petty maneuvers. Badly positioned his troops; Guan Yu flooded them and captured him. Yu Jin surrendered. After Lü Meng took Jing Province, he released the imprisoned Yu Jin back to Wei. Later scorned by Cao Pi; died in despondency.

Pang De: Previously under Ma Chao. Extraordinarily brave — a personal favorite. Carried his own coffin into battle. Could fight Guan Yu for 100 bouts — the highest honor in the solo-combat world. His reputation doesn’t match the Five Tiger Generals, Xu Chu, or Dian Wei, but I personally believe his solo combat ability is on the same level. Unfortunately, never truly utilized. Then dragged down by his deadweight teammate Yu Jin: Guan Yu flooded seven armies and captured him. Refused to submit to Guan Yu, refused to surrender, was executed. A true hero.

Li Dian: Frequently appears leading troops in the early-mid period. Captured Huang Shao alive. Solo combat: exchanged about ten bouts with Zhao Yun, realized he was outmatched, turned his horse and retreated. Never seen again after the Hefei campaign.

Lady Zhen: Yuan Xi’s wife. After Yuan Shao’s defeat, Cao Pi snatched her and made her empress.

Sima Yi: Late-period god-tier grand strategist. Can read stars. Can even grab a blade and solo. Fought Zhuge Liang to a standstill around Hanzhong. Never lost to Liang at the grand strategic level. Later seized Cao Shuang’s military power; the Sima clan took control of Wei.

Sima Shi: Sima Yi’s eldest son. His characterization in the late period is relatively well done. “Round face, large ears, square mouth, thick lips. Under his left eye grew a black mole, from which sprouted dozens of black hairs.” While battling Wen Yang, “his eyeball burst out from the mole’s wound, blood streaming across the ground. In unbearable agony, yet fearing it would unsettle the troops, he merely bit his quilt and endured — biting the quilt to shreds.” Then bedridden. Shortly after, “with a great cry, his eye burst forth, and he died.”

Sima Zhao: Sima Yi’s second son. Prince of Jin.

Sima Yan: Sima Zhao’s son. Emperor of Jin.

Deng Ai: Late-period undefeated war god. Fought Jiang Wei to a standstill, never lost at the grand strategic level. Rolled down a cliff wrapped in felt, launched a surprise raid into Shu — the Shu people thought divine soldiers had descended from heaven and opened their gates in surrender. The man who conquered Shu.

Shu Side
#

Liu Bei: Everyone says Liu Bei’s benevolence was fake — personally, I think that’s an embellishment. From the novel’s portrayal of Xuande, Bei was genuinely benevolent. If he’d just taken Liu Biao’s resources in Jing Province directly, none of that mess would’ve happened. When entering Shu, the outcome was indeed duplicitous, but the novel still portrays Xuande with benevolence — I choose to respect the original here.

Guan Yu: A wildly popular character. Personally not a fan (in real life, this kind of person is extremely annoying). Early period: a god. Cut down foes in a single stroke. Single-handedly drove back Xu Huang + Xu Chu (probably only Lü Bu could match that feat). Arrogant and rude. Bears the lion’s share of blame for the loss of Jing Province. The loss of Jing is the novel’s plot turning point — Cao Cao, Liu Bei, Zhang Fei, Huang Zhong all die in rapid succession; remaining generals and strategists all fade from the storyline. Also, this man has an arrow-magnet constitution: shot during Crossing Five Passes, shot by an “air arrow” at Changsha fighting Huang Zhong, shot fighting Pang De, shot with a poisoned arrow attacking Fancheng.

Zhang Fei: Zhang Fei is a highly stylized character, but his combat record is better than Guan Yu’s. “Round-eyed rogue,” brave and cunning, hates evil like an enemy, true to his nature. The only person in the entire Three Kingdoms who dares to taunt Lü Bu. Can fight Lü Bu for 100 bouts. Drank off Cao Cao’s army at Changban Slope. Marched into Shu by land. Honorably released Yan Yan. Shattered Zhang He. Stabbed and wounded Xu Chu. Can lead troops, can solo, has tactical intelligence — a top-tier Three Kingdoms general. One scene is especially moving: after Guan Yu’s death, Liu Bei kept delaying the revenge campaign. Zhang Fei said to Liu Bei: “Our brother is dead — what’s the point of being emperor?” “If you won’t avenge our brother, don’t bother seeing me again.” 👍. I previously visited Zhang Fei’s tomb in Langzhong — his calligraphy was remarkably refined, nothing like the crude brute you’d imagine…

Zhuge Liang: A monster.

Pang Tong: “Sleeping Dragon and Young Phoenix — obtain one and you can have the realm” is pure bluster. Cannot be ranked alongside Zhuge Liang. Combat record is basically negative.

Xu Shu: God-tier grand strategist. Under Liu Bei (attached to Liu Biao), engineered the first-ever defeat of Cao Cao’s army (Cao Ren). Defining trait: filial piety… Cheng Yu forged a letter from his mother to summon Yuanzhi. Xu Shu went to Cao Cao’s camp; after his mother committed suicide, Xu Shu, out of pride, still wouldn’t return to Liu Bei’s side… utterly baffling.

Fa Zheng: The strategist for Huang Zhong’s army when Xiahou Yuan was killed. Other schemes had no weaknesses. Died early. One of only two people Zhuge Liang ever sought advice from.

Ma Su: The other person Zhuge Liang ever sought advice from. During the Southern Barbarian campaign, he was the first to propose a strategy aimed at winning hearts rather than annihilation. As long as he didn’t lead troops himself, god-tier. First time leading troops: defeated by Sima Yi. Later executed by the Chancellor. Also: the Chancellor shedding tears as he executed Ma Su — he wasn’t crying for Ma Su, but lamenting that the late Emperor’s legacy of the northern expedition remained unfulfilled.

Zhao Yun: Never lost a solo fight. No one could go 100 bouts with him. Basically, he’d show up and “spear them dead in one thrust.” Evasion maxed out — “the hero of Changban Slope is still in his prime.” Rarely led troops; more like Liu Bei’s personal guard, protecting the imperial family. Liu Bei called him brother, but he never entered the core trio.

Huang Zhong: Fought Guan Yu for 100 bouts at Changsha. His horse stumbled and Guan Yu spared him. Later shot an arrow without the arrowhead attached, repaying the debt. Killed Xiahou Yuan in the Hanzhong campaign.

Ma Chao: “Splendid Ma Chao.” Cao Cao: “Ma Chao’s valor is no less than Lü Bu’s in his prime.” Nearly made Cao Cao cut off his beard and discard his robe in flight — Cao was saved by Cao Hong. Fought top-tier warriors Xu Chu and Zhang Fei for 200 bouts each. Rash and cruel; committed city massacres. Limited achievements under Liu Bei.

Wei Yan: Had “a rebellious bone at the back of his skull.” An important Shu general in the mid-late period. Solo combat: decent. Leading troops: first-rate. Zhuge Liang predicted that after his death, Wei Yan would rebel — killed by Ma Dai. The famous Ziwu Valley gambit: though Sima Yi praised it, I personally think it’s a bit far-fetched.

Yan Yan: Solo combat ability basically zero. Pummeled by Zhang Fei. Archery ability: top-tier — shot Zhang Fei’s helmet. Participated in the Hanzhong campaign. No further appearances.

Huang Yueying: Actually, not much of a role. Only described when introducing Zhuge Liang’s son Zhuge Zhan: “The mother was exceedingly ugly but possessed extraordinary talents: versed in astronomy above, geography below; there was no book of strategy, divination, or escape arts she had not mastered.”

Zhuge Zhan: Son of the Marquis of Wu (Zhuge Liang). Hyped up upon debut, then sent against Deng Ai — killed by Deng Ai.

Liu Feng: Liu Bei’s adopted son. Solo combat ability: low. No tactical sense. Easily persuaded. A net negative. Guan Yu disliked him. Later, when Guan Yu was defeated and sought reinforcements, Liu Feng and Meng Da refused to send troops, contributing to Guan Yu’s death. Subsequently executed by Liu Bei.

Meng Da: Betrayed, then betrayed again. Shares blame for Guan Yu’s death. Only highlight: shot and killed Xu Huang. Later killed by Sima Yi.

Liu Bei’s Wives: Lady Gan — the one who threw herself into the well at Changban Slope, A-Dou’s birth mother. Lady Mi — died while Liu Bei was in Jing Province, which led to Wu’s marriage proposal. Sun Shangxiang — a fifty-something old ox marrying a sixteen-year-old girl…

Mi Zhu: Brother of Liu Bei’s wife Lady Mi. A tool character — basically Liu Bei’s envoy for delivering messages.

Mi Fang: Brother of Liu Bei’s wife Lady Mi. Technically the Emperor’s brother-in-law, yet surrendered to Wu. Bears responsibility for Guan Yu’s death.

Sun Qian, Jian Yong: Followed Liu Bei in the early period. No particular talent. Tool characters.

Guan Ping: Guan Yu’s adopted son. Fought Pang De for thirty bouts. Later captured alongside Guan Yu by Wu; executed.

Guan Xing: Guan Yu’s biological son. A key general in the mid-late period. Killed Pan Zhang — his father’s murderer — and recovered the Green Dragon Blade. A main combat general on the Qishan campaigns. Later died of illness.

Zhang Bao: Zhang Fei’s biological son. A key general in the mid-late period. Appears alongside Guan Xing.

Liao Hua: Originally a Yellow Turban, later followed Guan Yu. During the desperate escape from Mai Castle, ran out to seek reinforcements and survived. Later appears on the Qishan campaigns.

Zhou Cang: Originally under Zhang Bao, later followed Guan Yu — carried Guan Yu’s blade. Fought Zhao Yun and lost repeatedly, taking three spear wounds. Solo combat: weak. Committed suicide after Guan Yu’s death.

Ma Dai: A late-period Shu general. Frequent appearances. Achievements in the Southern Barbarian campaign and Qishan expeditions. Under the Chancellor’s brocade-bag stratagem, executed Wei Yan.

Jiang Wei: A Wei defector. Inherited Zhuge Liang’s will. Accomplished in both letters and arms. Launched (I think ten) expeditions from Qishan… Later, Deng Ai raided Shu; Liu Shan surrendered. Jiang Wei was still holding Jianmen Pass… “We fight to the death — why do you surrender first!”

Wu Side
#

Sun Jian: Among the useless warlord coalition, capable of fighting. Obsessed with the Imperial Seal. Swore he didn’t have the Seal — “may I be shot dead by random arrows.” Later shot dead by Huang Zu’s troops.

Sun Ce: Sun Ce the Little Conqueror. A fierce warrior. Essentially conquered all of Jiangdong single-handedly — just died too young. A personal favorite. Trading the Imperial Seal for troops to build his kingdom — a stroke of genius, surpassing his father. Killed by an enemy’s revenge attack. While recovering, because he refused to believe in superstition, was mystically killed by Yu Ji.

Yu Ji: The people thought he was an immortal. Could cure people with talisman water. Executed by Sun Ce. His ghost haunted Sun Ce and killed him…

Sun Quan: Zero military talent whatsoever. Pummeled at Hefei. His strong point: recognizing talent. All four of Wu’s early (and most important) Grand Commanders were strong.

Taishi Ci: Appeared quite early — already present when Liu Bei was helping Tao Qian. Later joined Liu Yao, then was subdued by Sun Ce. Later, done in by Sun Quan at Hefei…

Gan Ning: “Brocade Sail Pirate.” Wu’s number one combat power. Expert archer. (Honestly, Wu’s generals’ combat ability is not impressive.)

Ling Tong: His father Ling Cao was killed by Gan Ning — a blood feud. During the Hefei campaign, saved by Gan Ning; they reconciled.

Huang Gai: Master of getting beaten. Actually, no combat highlights on the battlefield. At Red Cliffs, shot off his boat by Zhang Liao with one arrow, rescued by Zhou Yu. No further news.

Zhou Yu: Wu’s first Grand Commander. A winner in life. A personal favorite. Too bad: “Since Heaven gave birth to Yu, why did it also give birth to Liang?”

Zhang Hong: No role.

Zhang Zhao: Leader of the dove faction. Default answer to everything: surrender. Virtually none of his schemes ever worked.

Lu Su: Wu’s second Grand Commander. Timid. Appreciated by Zhou Yu. No combat achievements. Many say he was “shrewd beneath a foolish exterior” — the original does have hints of this, but “shrewd beneath foolish” is a stretch. Basically just a messenger between Liang and Yu. The originator of the empty-handed Jing Province recovery attempts.

Lü Meng: Wu’s third Grand Commander. Mastermind of “Crossing the River in White” (disguising troops as merchants). Can basically be considered the killer of Guan Yu. When he saw the beacon towers in Jing Province and couldn’t find a way to break through, he claimed illness and stayed home (absolutely hilarious). Later seen through by Lu Xun. After Guan Yu’s death, mystically killed by Guan Yu’s ghost.

Lu Xun: Wu’s fourth Grand Commander. Mastermind of Yiling. Later participated in several major campaigns. Lu Xun’s talent was not beneath Gongjin’s (Zhou Yu’s).

Zhou Tai: Fought his way in and out to rescue Sun Quan. For every wound he bore, Sun Quan made him drink a cup of wine.

Pan Zhang: Fought Guan Yu — lasted only three bouts before fleeing. Guan Yu’s spirit manifested; killed by Guan Xing.

Ding Feng: Shot and killed Zhang Liao. An important late-period Wu general. Survived until chapter 119.

Ma Zhong: Pan Zhang’s subordinate. Many have never heard of this character, but he killed both Guan Yu and Huang Zhong. Killing Guan Yu was cleaning up; killing Huang Zhong was genuine skill — one arrow took down the master archer Huang Zhong. Later assassinated by Mi Fang.

Jiang Qin, Han Dang, Xu Sheng…: Too many, unremarkable, can’t remember.

Sun Shangxiang: No such character exists in the official histories. In the novel, Sun Shangxiang has only personality description — she likes dancing with blades and swords. She never actually participated in combat. No children after marrying Liu Bei. Later tricked into returning to Wu; never saw Liu Bei again. But later generations adore Sun Shangxiang — she’s a fan-favorite character. Total War’s beauty icon:

Others
#

He Jin: General-in-Chief, Empress He’s brother. An utter fool. Held all the cards and played them terribly. To deal with the Ten Regular Attendants, he summoned Dong Zhuo to the capital, setting off an unstoppable chain reaction — the realm fell into chaos.

Zhang Jiao, Zhang Bao, Zhang Liang: Yellow Turban rebel leaders. Could cure with talisman water, summoned divine soldiers. The rest of the time, basically got beaten up by the regular army. A religious peasant uprising, hastily concluded. The novel dismisses it with “Zhang Jiao was already dead.”

Yuan Shao: Previously under He Jin — back then he was quite strategic, even dared to confront Dong Zhuo directly. “Many schemes but poor decisions.” His advisers each pushed their own agenda; none of his generals were worth anything.

Yuan Tan, Yuan Xi, Yuan Shang: Yuan Shao’s three sons. Still fighting over power after Yuan Shao’s defeat.

Lü Bu: The number one warrior of the Three Kingdoms. Early period: unstoppable in single combat. Only lost when ganged up on. (Late period Lü Bu once soloed Zhang Fei.) Cao Cao had suffered at Lü Bu’s hands — ultimately Cao Cao was the big winner.

Chen Gong: After Cao Cao’s failed assassination of Dong Zhuo, Chen Gong followed him — and witnessed Cao Cao’s treachery: “Better that I betray the world than let the world betray me.” Disgusted with Cao Cao, he left and later joined Lü Bu.

Zhang Song: Liu Zhang’s subordinate. Arrogant. Cao Cao disliked him. Later defected to Liu Bei, offered the map of Western Sichuan. Later discovered by Liu Zhang colluding with Liu Bei; executed.

Zhang Xiu: Featured prominently in early battles against Cao Cao. Originally surrendered to Cao Cao, but because his aunt was forcibly taken by Cao Cao, he rebelled. Cao Ang and Dian Wei died in this battle. Later defeated by Cao Cao again; surrendered.

Chunyu Qiong: Commander of the Wuchao supply depot. Drinking ruined everything.

Li Ru: Dong Zhuo’s strategist. Never appears again after Dong Zhuo’s death.

Zuo Ci: A full chapter of pure mysticism. Stunned everyone reading it — “Come out and see the immortal~”

Yu Ji (the physician): Cao Cao’s doctor. Tried to poison Cao Cao; discovered.

Hua Tuo: The Three Kingdoms’ number one physician. Skilled in surgery. Treated Zhou Tai’s wounds. Scraped Guan Yu’s bones to cure poison. Later, while treating Cao Cao’s head ailment, suspected of being a second Yu Ji — died in prison.

Chen Lin: One of the Seven Masters of the Jian’an period. His “Proclamation Against the Usurper” is recommended reading in full.

Final Thoughts
#

There’s simply too much to discuss. I thought I could finish this piece in two or three hours — ended up costing several times that. Romance of the Three Kingdoms is truly brilliant, absolutely worth reading (as if that needed saying). I probably won’t continue with the Records — time to urgently start the next chapter…

Book Notes — Space Odyssey Series

Mon, 12 Aug 2024 00:00:00 +0000

Preface
#

An unavoidable work for any sci-fi fan: Arthur C. Clarke’s classic — the Space Odyssey series. The Space Odyssey consists of four volumes: 2001: A Space Odyssey, 2010: Odyssey Two, 2061: Odyssey Three, and 3001: The Final Odyssey. As the titles suggest, the futuristic technological visions take place in their respective years. Don’t think 2001 has already passed — when Clarke wrote 2001, it was 1968! I, at least, can’t imagine what the world will look like thirty years from now, or how far humanity will have advanced in space exploration.

I wrote a reading reflection after first finishing 2001, captivated by its premise, its thrilling space plotlines, its fantastical cosmic backdrop… I immediately dove into the remaining three volumes. I initially expected the setting to expand ever outward, but that’s not what happened. The later three books remain within this cosmic dimension — between Jupiter and Earth — which is already very, very small. They mostly fill in plot details and imagination, bringing the entire story to completion.

The Tetralogy
#

2001: A Space Odyssey
#

After finishing the entire series, this one still feels like the most classic. Maybe it’s because the plot was the result of discussions with Kubrick…

Since I’ve already written a full reflection on it before, I won’t belabor it here. Interested friends can check out my earlier Book Notes — 2001: A Space Odyssey.

2010: Odyssey Two
#

This volume is also brilliant. In the original, China’s spacecraft sent to explore Jupiter is named the Qian Xuesen, and the Qian Xuesen is the first manned mission to land on and explore Jupiter’s moon — Europa — beating the Americans to it! Though the outcome wasn’t great, the plot is thrilling~ Before the Qian Xuesen’s accident, the astronauts described lower life forms on Europa and were ultimately attacked and killed by “extraterrestrial organisms.” This “disaster” plotline sparks infinite imagination: what kind of life exists on Europa? And what should we humans do about it?

Finally, the monolith on Jupiter goes through a series of self-replications and ultimately transforms Jupiter into a white dwarf! Jupiter is ignited! From then on, there are two “suns” in the sky. This premise is just fantastic~

2061: Odyssey Three
#

This one feels a bit rushed, mainly because Halley’s Comet was coming. Clarke wrote in the preface that since Halley’s Comet was about to sweep past Earth, if he didn’t release the book soon, the exploration-of-Halley plotline might become untimely. Indeed, a large portion of this volume is devoted to exploring Halley’s Comet. There are some Jupiter-related plotlines, but they don’t advance the main narrative much.

Halley’s Comet orbits the sun once every 76 years. Its next return is about 40 years away (July 28, 2061). Thinking back, its last perihelion was roughly when this book was written — the whole world was talking about Halley’s Comet. (I can feel that no one’s mentioned it in recent years.)

3001: The Final Odyssey
#

A perfect concluding work! This conclusion has influenced countless sci-fi novels — you can even clearly sense the shadow of The Three-Body Problem.

The first three volumes are all still in the 21st century. 3001 jumps a full thousand years! Humanity now acquires knowledge through “brain-computer interfaces” rather than learning; the speed of space travel has increased enormously…

But honestly, a thousand years — a thousand years and humanity has only progressed this far? I’d rather believe it was because of the sophons. Huh? Could it be that Old Liu’s sophons were inspired by this exact idea?

The most brilliant part of this volume is humanity resurrecting Poole — an astronaut killed by HAL in the first book. If no one brought him up, you’d assume he was still drifting in space… Resurrecting Poole not only echoes the first book’s plot but also allows us to observe and unveil the human world of the year 3001 through the eyes of an “ancient person.”

The Shadow of The Three-Body Problem
#

Or to put it the other way around: the shadow of Space Odyssey in The Three-Body Problem. I tried recalling from memory — apologies for any omissions:

The Sower — the Singer. In Space Odyssey, the Overlords are “planting” life; in The Three-Body Problem, the Overlords casually “eliminate” life — “What does it have to do with you?”
Alien warning. “Stay away from Europa” — “Do not answer! Do not answer! Do not answer!”
Display of alien technology. The Monolith — the Droplet. Both are materials beyond human comprehension, impossibly smooth, artifacts of alien civilizations that humanity’s technology cannot fathom. They represent the vast gap between human and alien technological levels.
Alien life is coming. We have time to catch our breath, but it seems like nothing we do will matter.
Resistance plans. With alien beings about to arrive, people begin formulating resistance plans. At this point, humanity still doesn’t know what the enemy truly looks like.

In fact, the two works differ greatly in many ways. Space Odyssey is about cosmic exploration, while The Three-Body Problem is about human society as a whole facing alien civilization. Space Odyssey essentially has only a handful of protagonists, even across a thousand years, and the plot mainly revolves around Jupiter. The Three-Body Problem has a grander scale and far more characters…

Final Thoughts
#

I’ve finally finished the Space Odyssey tetralogy. You can truly feel it’s a monumental work of science fiction — it satisfies a sci-fi fan’s longing for the “exploration” of space. Before any space agency had begun exploring “there,” Arthur C. Clarke had already arrived. NASA astronauts would even write back to Clarke: “We photographed the far side of the moon. There were no monoliths, no anomalies” — almost as if saying, “You fraud, I went there precisely because I read your book!” Haha~

Clarke wrote many plotlines about the Qian Xuesen spacecraft in the novels, and in the afterwords of several volumes, he repeatedly emphasized that Qian Xuesen was a person who profoundly influenced the aerospace industry — both in China and the United States. The U.S. arrested him on fabricated charges, and Qian Xuesen ultimately returned to his homeland to build its aerospace program from scratch, influencing missile development. During a trip to Beijing, Clarke even made a special attempt to visit Qian Xuesen, but at the time, Qian’s health was poor, and his doctors wouldn’t permit visitors. Clarke entrusted someone to deliver an autographed copy of Space Odyssey to Qian.

Reading the entire series, you can feel the era’s obsession with space exploration. But after the Apollo program was shut down, people seemed to lose interest in space altogether. However, with Musk’s Mars colonization plans, the theme of “space” seems to be returning to public consciousness. NASA says they’ll land on Mars by 2040 — who knows if it’s true. I’ll come back to dig up this post then.

Book Notes — To Kill a Mockingbird

Mon, 12 Aug 2024 00:00:00 +0000

Preface
#

Many people probably know To Kill a Mockingbird. I saw its ratings were sky-high and couldn’t resist picking it up. Sure enough, the story is brilliant — never a dull moment. Its style is quite different from the books I’d read before. Personally, I think it’s perfectly suited for middle school readers (no condescension intended) — a simple, fun, and superbly written story. In truth, the message the whole book wants to convey is very clear: don’t harm innocent people. The real difficulty lies in how to build a brilliant story around such a simple idea.

Prose Style
#

I think the best thing about Mockingbird is its prose. The author brings a small-town story in the American South vividly to life. Several storylines are stunningly rendered, the plot follows the timeline smoothly without feeling muddled, and it reads effortlessly and comfortably. The story plants foreshadowing from the very beginning, only unearthing the biggest reveal right at the end. The depiction of the gap between white and Black lives is also extraordinarily vivid. This book’s setting is contemporaneous with the HBO series Boardwalk Empire, which I’d watched before — that show also features Black neighborhoods, so I could easily picture the white and Black communities.

One scene where the protagonist gets beaten up left a deep impression: “I was pressed to the ground, and before my eyes was a tiny ant, laboriously hauling a breadcrumb through the grass.” I find it hard to articulate exactly what this passage means. She’s being assaulted, yet her attention is caught by an ant carrying a breadcrumb? Maybe it means nothing? But whatever the case, this description makes almost everyone mentally highlight it — it’s so visually evocative. And it feels very much like stepping out for a cigarette after being immersed in stressful work for too long… it yanks you from tense, urgent action into another quiet world, then immediately back again.

Another Prime Minister story also left a deep impression. The young protagonist asks her father: “What’s a ‘whore’?” Her father tells her a story about a Prime Minister blowing a feather: “Every day the Prime Minister sits in the House of Commons blowing a feather toward the ceiling, straining every sinew to keep it from drifting down — yet people around him keep losing their heads one after another.” Reading this, I was just as baffled as the protagonist. What on earth does any of this have to do with anything? I only figured it out after consulting Baidu. Her father meant: don’t obsess over irrelevant things. Which is to say, her father offered no explanation at all. But by the time of the rape trial, the young protagonist understood perfectly — she knew what “rape” meant. It’s hard to say whether this kind of evasive education is right or wrong.

Who Killed Bob Ewell?
#

The final chapters are brilliantly rendered. While reading, I felt completely immersed in that pitch-black schoolyard night, that son-of-a-bitch Ewell (that’s how the sheriff refers to Ewell in the book — the first time I read that line, I silently cursed him too…) hunting down two innocent children… In the end, Ewell dies, but the full truth of what happened isn’t entirely clear. The narrative is told from the young protagonist’s first-person perspective, but she doesn’t see who killed Ewell.

Since I read the e-book version, I could see many readers’ annotations and comments. I found that many people completely missed the key details of the case. I was also utterly confused after my first read-through. I reread the relevant sections several times and finally pieced together the author’s intent and the full sequence of events. Let me unravel this mystery through several key questions.

1) Is Boo Radley Black or white?

This question seems absurd but is critically important. If Radley were Black, then Tom Robinson’s case would be a cautionary tale — a Black man killing a white man is enough to be executed several times over. The jury wouldn’t care about the truth; the defendant being Black would be sufficient for a guilty verdict. So the old father and old sheriff’s desire to protect Radley would be perfectly natural — putting Radley through the legal process would just be throwing away a good man’s life. This would also make the novel a work primarily about Black racism.

But Radley is white. So none of the above applies. This also brings the novel’s content more in line with its title. The author never directly states that Radley is white, but you can put it this way: if the author doesn’t specify someone is Black, then they’re white~. Of course, there are other clues: Radley ran around with Cunningham boys (white) as a kid; he lives in a white neighborhood; his skin is deathly pale… Radley is a character described from the very beginning to the very end, the most richly drawn “mockingbird” of the entire book — yet we only see his true face in the final two chapters. That’s why I felt so unsettled not knowing whether he was white…

2) The gap in the action

Here is the passage where the young protagonist is pinned down by Ewell and ultimately saved:

“He was slowly choking me, and I couldn’t move at all. Suddenly, he was yanked hard from behind and fell to the ground with a thud, nearly dragging me down with him. I thought, Jem must have gotten up.

Sometimes, human reactions are sluggish. I stood there dumbly, like a mute. The sounds of struggle slowly subsided. Someone was panting heavily. The night returned to its prior stillness.

…I slowly realized there were four people under the tree now.”

From the moment Ewell is pulled away to the moment there are four people under the tree, a struggle took place. Afterward:

Jem (the protagonist’s older brother) lies on the ground, injured by Ewell, unconscious
Ewell (the man who tried to kill children) lies dead with a kitchen knife in his ribs
Radley (the man who came to save the children) leans against a tree, coughing
The protagonist stands frozen, still in shock

The “gap” refers to: who pulled Ewell away and killed him? What exactly happened? The subsequent discussion between Atticus and the sheriff revolves around reconstructing this gap.

3) The kitchen knife

First, the knife the sheriff uses for his demonstration is a switchblade, not the kitchen knife. “Was Ewell killed with this knife?” “No, that knife is still in him. From the handle, it’s a kitchen knife.” So the sheriff did not destroy the murder weapon — that’s a fact.

In any homicide, the murder weapon is an extraordinarily critical piece of evidence. Clearly, this kitchen knife is the murder weapon. Whoever brought this kitchen knife is very likely the killer. The sheriff says, “Ewell probably found that kitchen knife somewhere in the dump… sharpened it razor-sharp… Ewell fell on his own knife.” This is the sheriff’s subjective speculation. There are many possibilities:

Ewell brought the knife, tripped himself, and the kitchen knife stabbed into his ribs — an accidental death.
Ewell brought the knife; Jem, despite his broken arm, wrestled it away and killed him.
Radley brought the knife and killed Ewell.

First: the probability of Ewell bringing the knife is low. If he’d brought a knife, he could have just rushed up and stabbed them — there’d be no need to go to the trouble of twisting Jem’s arm and strangling the protagonist. Now:

Scenario 1: Ewell was already fighting someone (though it’s never explicitly stated with whom). An accidental death at this point seems far-fetched, but it can’t be entirely ruled out — though the probability is extremely low.
Scenario 2: Broken-arm Jem wrestles the knife away and kills Ewell. This scenario is based on the protagonist saying “it felt like Jem pulled Ewell back” — so naturally it must have been Jem fighting Ewell. The protagonist didn’t see who pulled Ewell back; she only says it “felt like” him. A thirteen-year-old boy with a freshly broken arm taking a knife from an adult and killing him — also an extremely low probability.
Scenario 3: Radley brought the knife, yanked Ewell back to stop him from strangling the protagonist, then stabbed Ewell to death. This is the most likely scenario — and precisely the scenario that Atticus and the sheriff “deliberately” avoid mentioning during their reconstruction. One detail supports this: before Ewell burst out, both children had screamed. The neighbors probably didn’t hear — but earlier in the book, it’s mentioned that the tree at the scene is very close to Radley’s house.

4) The reconstruction

Whether viewed through the novel’s themes and atmosphere, or through specific case analysis, it’s almost certain that the person who killed Ewell was Mr. Boo Radley.

The reconstruction dialogue between Atticus and the sheriff — I reread it multiple times; it’s absolutely fascinating. It traces the entire process of reasoning through Ewell’s death and Atticus’s and the sheriff’s psychological shifts — yet throughout, the real killer is never once named.

Atticus wants to clarify the facts; the sheriff wants to protect the child.

First, the protagonist says someone pulled Ewell away — she felt it was Jem. Based on this, Atticus deduces that Jem got up, pulled Ewell away, wrestled the knife from him, and killed him. Working from Atticus’s deduction that Jem is the killer, the sheriff wants to protect Jem and says, “Ewell fell dead on his own knife.” Atticus then says: “If we cover up the truth, that would go against everything I’ve ever taught Jem about how to be a person.” To convince Atticus, the sheriff even demonstrates the tripping scenario.

Gradually realizing the truth.

A thirteen-year-old boy with a broken arm is unlikely to fight and kill an adult in the dark. “Unless someone is very accustomed to the dark to qualify as a witness…” — an unmistakable hint at Radley, who never leaves his house.

The key piece of evidence — the kitchen knife.

Atticus “suddenly” asks about the knife. The knife is still in Ewell’s body. Both of them individually realize the knife is Radley’s. They need to smooth over the knife issue. The sheriff suggests maybe Ewell found it in the dump and sharpened it.

Confirming the lie.

There are many slow-motion descriptions interwoven here. Both Atticus and the sheriff are silently checking whether this lie has any holes, whether they should accept it: Ewell’s death was an accident — he killed himself. In the end, they reach an agreement. Even the eight-year-old protagonist says, “I can understand.”

The title drop.

“This man has done a great service for you and for this entire town. If people ignored his reclusive habits and forced him into the spotlight — I think, that would be a crime.” This line directly hits the novel’s theme. A mockingbird symbolizes an innocent, harmless person. Dragging Radley — this mockingbird — into the spotlight is a crime. It echoes what Atticus said earlier: “Remember, it’s a sin to kill a mockingbird.”

Final Thoughts
#

To Kill a Mockingbird is a relatively simple story, easy to understand — at least compared to some of the books I’ve read before. The stories of the mockingbirds in the book left a deep impression. It reminds me of the “Brother Long” self-defense case from a few years ago in China. Without Brother Long, it’s likely that killing someone in self-defense would still get you a prison sentence here. Think about America nearly a hundred years ago, when the law itself was still newly established… Think about that Black man wrongly convicted, shot dead by prison guards while trying to escape. What kind of despair must he have felt in that prison cell? He just wanted to be a decent person — and his life was suddenly cut short.

Let me close with a line from Atticus: “I think there’s just one kind of folks. Folks.”

Book Notes — When Breath Becomes Air & What Life Should Mean to You

Mon, 12 Aug 2024 00:00:00 +0000

Why Write About Two Books Together?
#

Normally I’d write separate pieces after finishing these two books, but I figured neither would yield all that much content. Although I’ve read a few English originals (and written about them), I clearly underestimated the difficulty of When Breath Becomes Air. It’s packed with unfamiliar vocabulary — loads of medical terms I’d never encountered. I basically forced my way through it with half-understanding. As for What Life Should Mean to You… it doesn’t feel as miraculous as people say. After all, it’s a century old — I didn’t extract much nourishment from it (a little, though). To avoid the awkwardness of too-thin content, I’m lumping them together.

When Breath Becomes Air
#

The author was a surgeon with extraordinary achievements in medicine. At the peak of his career, he learned he had terminal cancer. Less than two years after the diagnosis, he passed away. This book was written during those two years. It describes, from a first-person perspective, how one confronts such misfortune as cancer and reflects on life and its meaning in one’s final days.

When he learned he had terminal cancer, as a top surgeon he knew exactly what it meant. He knew he didn’t have long to live. At first, he was even angry — why did such a low-probability event happen to me? Why me? Something like this is hard for anyone to accept. But only the truly ill live day by day with the pain, quietly walking toward that inevitable but unscheduled death.

After his diagnosis, he and his wife decided to have a child immediately — before chemotherapy began. The author also managed to spend a few months with his baby daughter before passing. He wanted to watch his precious daughter grow up, to know what she’d be like when she was older — though he was certain he’d never know. It seems almost too cruel.

Near the end of the book (about twenty or thirty pages from the finish), the author’s prose abruptly stops. What follows is a chapter written by his wife, opening with: “Paul has left us…” Even knowing how it would end, I couldn’t accept it — death came so suddenly that he couldn’t even finish his book… But thinking about it from the book’s intended meaning, this incompleteness is, in a way, a kind of completion…

How should we view death? If I were to die before forty, what would I do? I’d certainly be unwilling — there are too many things I haven’t finished. The author ultimately saw through the meaning of life; he believed the most important thing is to experience life and live in the present moment. I seem to be different — I live in the future, never now! If I go die right now, I’d leave this world accompanied by anger and resentment.

(His experience inevitably reminds me of the Japanese drama The White Tower — an absolutely brilliant show! Professor Zaizen, at the peak of his career, gets cancer and ultimately donates his body for cancer pathology research…)

What Life Should Mean to You (Beyond Inferiority)
#

A famous work in psychology by Alfred Adler, founder of individual psychology. Long ago, I watched an episode of Lao Gao and Xiao Mo about Adler and individual psychology — they made it sound almost miraculous. I couldn’t resist reading it, and figured I might even analyze myself a bit.

The most important idea in individual psychology is: how we perceive traumatic experiences is the essence of psychological problems — not the experiences themselves causing them. But this doesn’t deny the influence of the “past” on people’s behavior.

However, I personally found the book somewhat boring… “a bit too humanities-oriented.” The essential differences between short chapters aren’t that significant — it’s just discussing individual psychology through different topics. I genuinely couldn’t extract substantial nourishment from it. Maybe because it was written a hundred years ago, or maybe I’m just not cut out for this.

Adler also proposed: a group (or a couple) should think and act for the benefit of the collective, or else problems of separation will arise. If one person harbors self-serving thoughts, the group is bound to be unstable. I couldn’t agree more. I could write some self-analysis here, but I don’t want to expose myself — which is also why I felt this book note wouldn’t be very substantial.

Before reading this book, I also sampled The Courage to Be Disliked and How to Win Friends and Influence People. Since both had higher ratings than the “founding father” Adler’s book, I checked them out to see what they were about — and I didn’t like either. Courage is just a dialogue between two people — the classic wise-man-and-scholar format — where you learn the book’s ideas through conversation… I gave up after a bit. Bestseller style. How to Win Friends was more tolerable — it directly lays out life advice in plain terms. I read about ten pieces of advice — somewhat valuable — but I still couldn’t finish it. Bestseller style too.

Closing
#

I’d read a few English originals and clearly got a bit cocky — turns out I need to be realistic. Gauge the difficulty first before diving in. I’d been wanting to read psychology for a while. After reading it, I’ve learned I’m not cut out for it. Well, no matter what, I had to write this book note — recording my life, like Paul did.

Book Notes — Wild: From Lost to Found on the Pacific Crest Trail

Mon, 12 Aug 2024 00:00:00 +0000

Preface
#

I came across this book because I saw it on an Obama-recommended reading list in an e-book app. This particular title felt special, and the ratings were good, so I decided to check it out. At first, reading the synopsis — a memoir by a hiking enthusiast — I assumed the book would just describe scenic views and the hardships of sleeping rough, probably not very “exciting.” But its writing has a distinctiveness all its own; it never feels boring. Once you start a short chapter, you simply can’t stop. By the end, when I saw only 10% of the pages remained, I actually felt a sense of imminent parting — a reluctance to say goodbye. This feeling of having discovered a treasure accompanied me throughout the entire reading.

I used to be a devotee of physical books — I liked the sense of weight and substance, and the satisfaction of finishing a paper volume. Later, as I gradually came to embrace e-books, I discovered one advantage e-books have over physical ones: links. I found this book because a book mentioned in Space Odyssey (I think) led me, via links, to “Obama’s recommendations,” and from all the Obama-recommended books I picked a few that interested me — one was Wild. The protagonist of Wild also loves to read, and she mentions several books; I bookmarked about five or six of them. So my originally barren reading list grew and flourished through this chain of links. These books are far, far better than those “Top Book Rankings” or “Essential Classics, Domestic and International.” Finishing a physical book easily leaves you wondering what to read next; e-books don’t have that problem.

The Queen’s Journey
#

After losing her mother, seeing her family fall apart, facing massive college debt, and descending into drug addiction, the author — perhaps wanting to rediscover herself — made “thorough” preparations and set out for the Pacific Crest Trail. For someone with zero hiking experience, the Pacific Crest Trail is the highest difficulty level. She called her overloaded backpack “the Monster” — so heavy she couldn’t even put it on properly. And just like that, this outdoor novice set off. Completing the entire trail takes four to six months. Along the way, you need to plan resupply points in advance, mailing food and essential supplies ahead to those locations. Once you reach a resupply point and restock, you return to the trail and press onward. The suffering on the journey — though hard to feel vicariously — you can sense how severe it was. The author alone had six toenails removed. This kind of agony, along with various unexpected incidents, is beyond what the average person can endure. That’s why the “failure rate” for people attempting this trail is very high. You need an exceptionally robust physique, a thorough plan, and some luck.

Even with the dangers of wild animals, venomous snakes, scorching sun, glaciers, dehydration, and injuries in the wilderness, none compare to the danger of “people” — especially for a solo woman in her twenties. Once you experience the potential threat posed by humans, nature’s objective dangers almost feel like a relief. This reminds me of the plot of the HBO series The Last of Us, which I watched recently: in a post-apocalyptic world, encountering zombies isn’t the scariest thing — encountering humans is.

The author seems somewhat lascivious (at least by sexually conservative standards) — or maybe all Americans are this open about it. Before hitting the trail, she’d have one-night stands with many men, thoroughly enjoying the feeling, unapologetically describing this physical need and the sense of conquest in capturing a man.

What I craved wasn’t someone to love, but just someone to press their body against mine.

On the trail, she also fantasized about attractive men she encountered and secretly watched men undress. She also packed a lot of condoms in her backpack — sadly, not a single one was used by the end. Of course, I don’t mean she was only uninhibited about physical desires; she was just as emotionally and sentimentally passionate. No right or wrong — she simply expressed exactly how she felt in the moment. I really admire this kind of honest writing.

The Pacific Crest Trail
#

The Pacific Crest Trail is one of the world’s famous long-distance trails, located in the mountain ranges of the western United States — a range jokingly called “America’s Dragon Vein”… Trail information is very easy to find and extremely well-documented. The author also relied on trail guidebooks for preparation and handling unexpected situations. The trail stretches 4,000 kilometers, spanning the contiguous United States from the Canadian border to the Mexican border, passing through Washington, Oregon, and California. It is one of the National Scenic Trails.

I have essentially zero contact with hiking — my concept of it is still nil — so I can only be an armchair traveler envying these backpackers. After a bit of searching on outdoor hiking, I found there’s a tremendous amount to learn. Outdoor hiking not only offers spectacular scenery but apparently even has therapeutic functions — I easily found hiking psychotherapy associations just by searching. Let me quote a passage from the original about the spiritual world on the trail:

Now, I was wholly immersed in this world, living in a completely new way. Living so rootlessly, without even a roof over my head for shelter from wind and rain, made the world both much larger and much smaller.

The Golden Touches
#

While reading this book, I kept thinking of Educated: A Memoir. Both are memoirs describing a period of the authors’ pasts. Not only are their writing styles similar, but their upbringings are too — growing up in isolated mountains, having an abusive father, a backward family life, and unexpectedly being extremely good at studying and getting into a top university.

More importantly, their writing style easily captures the reader’s emotions without ever feeling boring or stifling. I haven’t managed to fully summarize how they write, but one thing I paid special attention to: the accumulation of emotion followed by the unexpected move.

For example: when the author scatters her mother’s ashes into the earth, she keeps a few larger fragments of bone, unable to let go. Finally, she puts these unburned bone fragments into her mouth and swallows them.

I was stunned reading this. Throughout the book, she describes her feelings for her mother in many places. Her mother’s death affected her profoundly. After flatly (or perhaps despairingly) describing her mother’s death and cremation, unable to let go, she chooses to swallow her mother’s bones into her stomach — so she can become one with her mother! What kind of emotion could drive such an act — one that most people would find impossible to accept — as a vessel for such heavy feeling? This swallowing motion conveys far more powerfully than endlessly repeated expressions of longing ever could, and it grabs the reader’s attention far more effectively.

There’s also a passage about condoms. An older backpacker, seeing how much stuff she’s carrying, helps her sort through her pack, throwing out things that are completely useless. The old backpacker finds a big packet of condoms: “Are you sure you need these?” Having gained some trail experience, she knows the stuff is utterly useless — but as she throws out the big pack, she secretly keeps one~ Then, the next morning when she wakes up, that one condom is gone…

These plot points are so dramatized I almost suspected they were fabricated. But I carefully read the author’s preface — she says she merely omitted certain scenes and guarantees that the events are all true.

Regardless, a touch of plot that slightly exceeds realistic logic is essential in writing — it grabs the reader’s heart. The authenticity of these “golden touches” themselves isn’t important; what matters is having that touch. Let me give an example from one of my favorite films, Memories of Murder, which I’m sure many have seen. Years later, the old detective returns to the crime scene and meets a child. The child says someone else was just here, crouching and staring at this drainage ditch just like you. The old detective immediately realizes this person could be the murderer. He asks the child what the person looked like. The child says: “Just… ordinary.” This moment is a stroke of genius. Many viewers obsess over who the killer actually is, but it doesn’t matter who it is. “The murderer is ordinary” — that’s what the film is trying to say.

Final Thoughts
#

What could have been a boring story was written into a captivating work, with genuine depth, rich and authentic emotion. It’s a memoir of following the author back to nature and rediscovering the self — absolutely worth reading!

Recently, good books have been streaming in nonstop; my bookshelf is quite packed. But I’m not worried about them gathering dust at all, because I believe the quality of these books matches this one — reaching that “can’t-put-it-down” level, requiring no self-discipline to become completely immersed. Let me quote from a hiking expert’s blog:

> What was your favorite stretch of scenery? > The next one.

读书笔记 on Last DBA

Book Notes — Designing Data-Intensive Applications (2nd Edition)

Ch1: Trade-offs in Data System Architecture #

OLTP & OLAP #

Cloud Services vs. Self-Hosting #

Cloud-Native #

Operations in the Cloud Era #

Ch2: Defining Non-Functional Requirements #

Hardware and Software Faults #

Scalability Principles #

Operations #

Ch3: Data Models and Query Languages #

SQL & NoSQL #

Ch4: Storage and Indexing #

Hash Indexes #

B-Tree Indexes #

SSTables & LSM Trees #

In-Memory Databases #

Materialized Views and OLAP #

Column-Oriented Storage #

Ch5: Encoding and Evolution #

REST vs. RPC #

Ch6: Replication #

Multi-Leader Replication #

Ch7: Partitioning #

Range Partitioning and Hash Partitioning #

Ch8: Transactions #

ACID, BASE #

Pessimistic and Optimistic Transaction Models #

Ch9: Distributed Systems #

Clocks #

Ch10: Consistency and Consensus #

CAP #

Distributed Transactions and Consensus #

Ch11: Batch Processing #

MapReduce and Distributed File Systems #

Hadoop vs. Distributed Databases #

Diversity of Processing Models #

Ch12: Stream Processing #

Event Sourcing #

Ch13: The Future of Data Systems #

Lambda Architecture #

Separation of Application Code and State #

Problems Introduced by Algorithms #

Summary #

Book Notes — Average Is Over

Why I Read This Book #

Chess #

Polarization #

The Rising Cost of Learning #

Self-Learning #

Finally #

Book Notes — Educated and Atomic Habits

I Hate Writing Essays #

Reading Originals Is Good for Body and Mind #

Educated #

Atomic Habits #

Finally #

Book Notes — Elon Musk

Gifted #

Arriving in America #

Zip2 and PayPal #

SpaceX and Tesla #

For Humanity #

Book Notes — Rich Dad Poor Dad

Started Reading E-Books #

Poor Dad and Rich Dad #

Finally #

Book Notes — Sapiens: A Brief History of Humankind

The Cognitive Revolution and Fiction #

Money #

Columbus and Zheng He #

Closing #

Book Notes — 2001: A Space Odyssey

God-Tier Predictions #

Imagination #

Old Liu (Liu Cixin) #

The Film — 2001: A Space Odyssey #

Closing #

Book Notes — Are We Smart Enough to Know How Smart Animals Are?

Ch1: Trade-offs in Data System Architecture
#

OLTP & OLAP
#

Cloud Services vs. Self-Hosting
#

Cloud-Native
#

Operations in the Cloud Era
#

Ch2: Defining Non-Functional Requirements
#

Hardware and Software Faults
#

Scalability Principles
#

Operations
#

Ch3: Data Models and Query Languages
#

SQL & NoSQL
#

Ch4: Storage and Indexing
#

Hash Indexes
#

B-Tree Indexes
#

SSTables & LSM Trees
#

In-Memory Databases
#

Materialized Views and OLAP
#

Column-Oriented Storage
#

Ch5: Encoding and Evolution
#

REST vs. RPC
#

Ch6: Replication
#

Multi-Leader Replication
#

Ch7: Partitioning
#

Range Partitioning and Hash Partitioning
#

Ch8: Transactions
#

ACID, BASE
#

Pessimistic and Optimistic Transaction Models
#

Ch9: Distributed Systems
#

Clocks
#

Ch10: Consistency and Consensus
#

CAP
#

Distributed Transactions and Consensus
#

Ch11: Batch Processing
#

MapReduce and Distributed File Systems
#

Hadoop vs. Distributed Databases
#

Diversity of Processing Models
#

Ch12: Stream Processing
#

Event Sourcing
#

Ch13: The Future of Data Systems
#

Lambda Architecture
#

Separation of Application Code and State
#

Problems Introduced by Algorithms
#

Summary
#

Why I Read This Book
#

Chess
#

Polarization
#

The Rising Cost of Learning
#

Self-Learning
#

Finally
#

I Hate Writing Essays
#

Reading Originals Is Good for Body and Mind
#

Educated
#

Atomic Habits
#

Finally
#

Gifted
#

Arriving in America
#

Zip2 and PayPal
#

SpaceX and Tesla
#

For Humanity
#

Started Reading E-Books
#

Poor Dad and Rich Dad
#

Finally
#

The Cognitive Revolution and Fiction
#

Money
#

Columbus and Zheng He
#

Closing
#

God-Tier Predictions
#

Imagination
#

Old Liu (Liu Cixin)
#

The Film — 2001: A Space Odyssey
#

Closing
#

Preface
#

Apes
#

What Makes Us Different?
#

Closing
#

Preface
#

Power and Alliances
#

Males and Females in Power
#

Power and Sex
#

Closing
#