Story 1: From Capitalism to Compute-ism

Company as a Compressor

If you’ve read my previous post Compression is Intelligence, the chain of equality should be crystal clear: Company = Compression, Compression = Intelligence.

In the era of classical Capitalism, a company was essentially a pricing mechanism for “intelligence.” To borrow Jack’s thought experiment: The boss uses capital to buy out employees’ bandwidth, and through a rigid hierarchy, “compresses” chaotic labor and information into standardized output.

Here, compression is a compromise for the low-bandwidth limitations of carbon-based life forms. The inverse process is the expression of world knowledge. A company’s market cap is simply the digital expression of “the intelligence this compression can restore.”

But now, the Hard Fork has happened.

When LLMs took over the interpretation rights of “intelligence” with unprecedented efficiency, coldly obeying the scaling law, there is only one logical endpoint: The dissolution of the company.

Transformers compress information far more efficiently than administrative orders. In traditional companies, those endless meetings, polished PPTs, and layers of reporting are essentially using expensive carbon-based bandwidth to simulate cheap silicon-based bandwidth (GPU communication).

Today, this “simulation” looks absurdly inefficient. Traditional management layers have turned into high-impedance nodes in the flow of information.

You can see this in every top LLM lab: Organizations are flattening rapidly. Decision-makers are bypassing layers of middle management to sync cognition directly with soldiers in the trenches. Because in the era of Compute-ism, information doesn’t need “humans” as a compression relay — the model itself is the most efficient medium.

If the org chart fails, what does market cap represent? It still represents “restorable intelligence.” But today, this compression no longer relies on headcount, but “GPU-count.”

Compute is the Gold Standard of the AGI era; Intelligence is the only currency.

This implies that capital once used to build massive orgs is becoming surplus. There’s far more capital “missing the bus” than there are investment opportunities. Once existing capital cannot convert into effective compute, it accelerates depreciation and physically evaporates.

Capitalism hasn’t vanished; it just swapped its host from “Humans” to “Compute.” And in this Hard Fork of host transfer, individual destiny is being rewritten.

Story 2: The AGI Era is the Era of Super Individuals

Individual as Leverage

When the logic of the “Company” compressor is rewritten, the flip side is the rise of the Super Individual.

Old-school influence was strictly bounded by biological stamina and physical time. To scale, you had to hire people and leverage their labor. This meant getting “alienated” by the organization, becoming a prisoner of the management chain, burning energy on communication and coordination.

But in the AGI era, individuals plug directly into compute leverage. With APIs and Agents, one person’s taste, judgment, and decision-making can be infinitely amplified:

• No need to suck up to VCs who don’t get the story;
• No need to babysit employees full of human contradictions;
• As long as you have the compute, you can instantiate 10⁴ LLMs in one second.

These silicon-based employees are united, need no motivation, need no “vision-selling,” and sustain peak output 24/7.

Capital flow will shift drastically: from “Chasing Organizations” to “Chasing Nodes.” An individual with high compute utilization potential becomes a highly efficient, iterative carrier of intelligence.

For the first time in history, an individual’s “Intellectual Creativity” and “Reality Creativity” are linearly aligned. Between the idea in your head and its construction in reality, the long engineering cycles and organizational friction are gone.

This explains the “NBA-style” massive transfer fees we see today. When a single brain (Node) can leverage infinite compute (Leverage), their valuation stops following traditional salary bands and starts following Asset Pricing logic.

Story 3: Curiosity Shapes the Super Individual

Curiosity Is All You Need

In the compute era, winning isn’t about how much compute you have (scalar), but where you point it (vector). Unless you own a proprietary cluster, for most carbon-based agents, the only optimization space is the in-context learning efficiency of the brain — our primitive compressor.

The curiosity I define isn’t a mild desire to learn, but an optimization strategy to resist collapse and pursue extreme entropy gain.

3.1 Hardware Constraints: Outsourcing is the #1 Productivity Force

Reference: Knowledge Storage and Extraction

Physics of Language Models Part 3.1 theoretically defines the storage limit of Transformers: capacity scales linearly with parameters. The human brain is likely the same.

The Super Individual’s first principle: Outsource. Offload all standardized knowledge to LLMs. Free up precious biological storage for that incompressible, undigitized “Dark Knowledge.” Once dark knowledge is tokenized into the training set, it becomes a commodity — outsource it immediately.

As they say in the Valley: “Knowledge is commodity; Insight is scarce.”

3.2 Input: Data Cleaning is Comedy — Violation of Expectation

Reference: Rho-1

Learn from Rho-1: Not all tokens are what you need. Reject zero-gradient information. If content is easy to read, drop it. These samples are in the center of your brain’s knowledge distribution; your neural weights are getting zero updates.

Actively sample edge cases at the boundaries of your cognitive distribution. “Life begins at the end of your comfort zone” is the only virtue in the AGI era.

3.3 Output: The Human Advantage is Pulling the Plug

Reference: The Second Half*, *CuES

Shunyu Yao mentioned in The Second Half that value is shifting from problem-solving (inference) to physical world alignment (real-world utility). A Super Individual’s output isn’t an answer, but a force that perturbs the environment. But CuES notes that tasks linked to the physical world are scarce (task scarcity).

LLMs can generate infinite plans digitally, but can’t verify which ones are feasible in reality (affordances). Carbon-based agents lose on raw thinking power but win on having bodies and the ultimate interpretation right — Pulling the Plug.

Precisely because of this, we are the only ones with “Skin in the Game” in the physical world. So, getting your hands dirty — I call it “Move fast and break things (online grounding).”

3.4 State Maintenance: Scratching Your Head is a Sign of Growth

Reference: CDE

Long-term collaboration with LLMs transmits their bad habit: confidently talking nonsense. It’s like hallucination, but hallucination is making things up. This is calibration collapse — the LLM is supremely confident it has converged to the truth.

To resist this IQ degeneration, two intuitive methods:

• Reward Perplexity: Confusion breaks security, but high perplexity means prediction failure. Awesome — this is an exclusive moment for efficient weight updates.
• Reward Variance: Be wary of consensus. Seek out high variance and controversy. This is the most hardcore regularization against overfitting.

Reading this, you’ll find the character profile is: Intellectually lazy (loves outsourcing), Obsessive (edge sampling), Action-oriented (physical verification), and Contrarian (seeking variance).

Sometimes, I feel that we — as a couple — are truly Curious Babies — destined LLM Researchers.

Outro

So the title of this post should have been How to Become Part of AGI, but for clickbait purposes, I shamelessly exaggerated. This is a typical out-of-distribution but imperfect case — an attempt to forcibly create uncertainty in a deterministic world.

If you have a better solution, or a more ferocious curiosity, please leave your masterpiece. I will seriously study it and attempt to “compress” your intelligence into my model.

Thanks for this efficient exchange of thoughts. Merry Christmas & Happy New Year!