A thesis · 2026

There are exactlyfour problemsin computing.

Every hard problem in computing is one of four problems — often two or three at once.

02 — Four Deep Dives

What each primitive really is.

1 / 4 — COMPRESS

Compression

More into less, without losing meaning.

Compression is the conversion of mass into representation. A model is a compressed corpus. A proof is a compressed argument. A hologram is a compressed scene. Every act of cognition is compression: a billion sensations become one thought.

01
Compression

Encoding — the original compression

Take one alphabet, rewrite it in another that's denser, more error-tolerant, or shaped to the channel. Huffman, ASCII, UTF-8, JPEG, AAC, H.264. Encoding is where compression first becomes engineering.

02
Compression

GPT — compressing the corpus

A trillion documents become a few hundred gigabytes of weights. Decompression is not retrieval; it is sampling. The model is lossy in detail, lossless in style, and learned how to be both.

03
Compression

Zero-knowledge proof — compressing an argument

A million-step computation becomes a few hundred bytes of proof — verifiable in milliseconds, revealing nothing more than that it ran. Compression and privacy in one stroke.

04
Compression

Holography — compressing a scene

An entire light-field encoded as a flat interference pattern, decodable from any angle. The original 3D-into-2D, with parallax recovered on read.

2 / 4 — ORDER

Ordering

Out of many, a sequence.

Ordering is the conversion of concurrence into sequence. Which transaction came first? Which fact matters most? Which item to show? Civilization runs on agreed-on orderings — and most civilizational arguments are over them.

01
Ordering

Transaction ordering — the consensus problem

Many parties broadcast actions; the system must produce one canonical sequence that all of them accept. This is what a blockchain actually is. Money is an opinion about ordering.

02
Ordering

Event ordering — what came before what

Lamport clocks. Vector timestamps. Causal graphs. Distributed systems can't observe absolute time, so they construct partial orders out of who-saw-whom. Time is something a system has to earn.

03
Ordering

Search — ordering by relevance

PageRank turned the web into a directed graph and asked: which page does everyone else point to? Search engines don't find pages. They sort them.

04
Ordering

Recommenders — ordering by you

Same problem as search, with the query replaced by an inferred you. The feed is a stream-shaped ranking. Attention has become a sorting workload.

3 / 4 — CONCUR

Concurrency

Many things at once, no collapse.

Concurrency is the art of letting many do work without stepping on each other. Hardware parallelism, distributed protocols, sharding, edges — all variations of "how do n things share one world?". The hard part is not speed. It is consistency under load.

01
Concurrency

Parallelism — one machine, many threads

GPUs, SIMD, vectorisation, async runtimes. Within a single box, find work that doesn't depend on itself and do it all at once. Amdahl's law says the floor is the sequential part.

02
Concurrency

Distribution — many machines, one promise

Replicate state across nodes; pretend it's one machine. The CAP theorem is the price tag: pick two of consistency, availability, partition-tolerance.

03
Concurrency

Sharding — divide the world to scale it

If state is too big for one node, chop it into independent slices keyed by user, region, hash. Cross-shard transactions are where the dragons live.

04
Concurrency

Edge — concurrency in space

Move computation toward the user. Latency falls, sovereignty rises. The cloud was a pendulum and it's swinging back to the device.

4 / 4 — PRIVACY

Privacy & Security

Choose what to reveal.

Privacy is the engineered right to reveal selectively. Cryptography is the toolbox. Symmetric keys hide content. Asymmetric keys distribute trust. Zero-knowledge proves without revealing. MPC lets adversaries compute together. All are answers to one question: who is allowed to know?

01
Privacy & Security

Symmetric crypto — one key, two parties

AES, ChaCha. If both sides already share a secret, you get speed and confidentiality. The unsolved half is how the secret got there in the first place.

02
Privacy & Security

Asymmetric crypto — whispering in public

RSA, elliptic curves. Two keys; one shouted, one whispered. The mathematical impossibility of eavesdropping. Every secure connection on Earth starts here.

03
Privacy & Security

Zero-knowledge — prove without telling

Prove you know the password, the path, the balance — without revealing it. The same primitive that compresses a long computation also conceals its inputs.

04
Privacy & Security

MPC — adversaries computing together

n parties each hold a secret; together they compute a function over those secrets without any one of them learning the others. Useful when there is no trustworthy host.

03 — The Composer

Pick a system. See its recipe.

Most real systems are recipes — two or three primitives stacked together. The size of each square tells you how central the primitive is.

Pick primitives
Psy Protocol
signature 12

ZK-native L2: compression and privacy are the same lever, ordering is its sequencer, concurrency is its sharded prover.

Compression
Ordering
Concurrency
Privacy & Security
ZK rollup
signature 10

Compress an L2 block's state-transition into one verifiable proof. Three primitives in one envelope.

Compression
Ordering
Concurrency
Privacy & Security
Bitcoin
signature 8

Order on top of crypto on top of concurrency. Money as an opinion about sequence.

Compression
Ordering
Concurrency
Privacy & Security
Google Search
signature 8

Compress the web into an index, sort it by relevance, serve it with massive concurrency.

Compression
Ordering
Concurrency
Privacy & Security
Ethereum
signature 7

Bitcoin's idea plus a programmable ordering — every smart contract is a clause on the sequence.

Compression
Ordering
Concurrency
Privacy & Security
ChatGPT-class chat
signature 7

Compress the corpus into weights, sample tokens in order, run batches of them in parallel.

Compression
Ordering
Concurrency
Privacy & Security
Git
signature 6

Content-addressed compression of file states, plus a causal ordering of commits.

Compression
Ordering
Concurrency
Privacy & Security
DNS
signature 6

A distributed naming service. Concurrency at planetary scale, with a hierarchical ordering of trust.

Compression
Ordering
Concurrency
Privacy & Security
Tor
signature 5

Privacy through layered encryption, plus the concurrency of a volunteer network.

Compression
Ordering
Concurrency
Privacy & Security
HTTPS / TLS
signature 4

Asymmetric crypto bootstraps a symmetric session. The web's privacy layer in two acts.

Compression
Ordering
Concurrency
Privacy & Security
04 — The Argument

Why only four?Because everything elseis plumbing.

Pick any hard system you can think of. Strip away the framework du jour, the cloud provider, the language. What is left? It will be one of these four shapes — usually a few of them stacked. Compression turns mass into representation. Ordering turns concurrence into sequence. Concurrency keeps the sequence from collapsing under load. Privacy controls who is allowed to read it. Everything else — the wire formats, the container runtimes, the build tools — is plumbing. Important plumbing. But still plumbing.

— Gewenbo