AI is a machine of probabilistic permutation is a thesis: a large model does not “retrieve” a ready-made answer from some database; rather, given the preceding text, it predicts and samples the next token by probability — at bottom, it is permutation and recombination of the knowledge humanity has already accumulated. From this, three conclusions follow. It can fill in the blanks of the phenomenal layer of human sensation at an exponential rate, yet it cannot inspire itself; it has no native “I don’t know” mechanism, and so it must produce hallucinations; and it amplifies the person already there without ever building a new dimension. The most condensed statement of the thesis is this: “AI is the summary of this age — through instructions it makes a probabilistic match and then fuses with the real world… AI can indeed inspire human beings, but AI itself is never inspired.”
Generation, Not Retrieval
The technical foundation of the thesis is a judgment about how the Transformer works: the model does not consult a lookup table, and there is no mapping in which “one question corresponds to one fixed answer.” The original wording is this:
The model is not retrieving from some database — it is doing probabilistic generation… There is no “correspondence” relation. There is no one question that corresponds to one fixed answer.
To mistake generation for retrieval is the most common misunderstanding of AI, and it is the source of every overestimation and disappointment that follows. Since each output is simply “given the preceding text, predict the most likely next word,” the entire boundary of AI’s capability is drawn jointly by the corpus it has studied and this probabilistic mechanism. This judgment springs from the same line of thought as the claim in Gaps in Cognition and Information: Can AI Replace the Economist? that “what AI processes is what has already been spoken.”
No “I Don’t Know”: The Architectural Root of Hallucination
The provenance of hallucination follows directly from probabilistic generation. Architecturally, the model is missing a native “stop”:
It has no native “I don’t know” mechanism. Its architecture dictates that it will always produce the next token with the highest probability — even when that probability is itself very low, very uncertain. This is one of the roots of hallucination.
This relocates hallucination from “an occasional error” to “a structural inevitability”: when real information runs short, the model does not fall silent or leave a blank, but spits out the highest-probability word all the same. In other words, it lacks the capacity to admit ignorance. This stands in sharp contrast to the way a person, meeting a boundary, can be “silent for want of focus” — see Silent for Want of Focus: Expression Lags Behind Depth; a person’s silence is honesty, while AI’s “refusal to be silent” is dictated by mechanism.
It Can Only Fill the Paper, Not Enlarge It
Pulling the lens back from a single output to the whole map of cognition, the thesis offers a central metaphor: the entirety of human knowledge is a sheet of “paper” already covered in lines (that is, the phenomenal layer), and what AI does is fill in every blank point on this drawing at an exponential rate — yet it cannot make the sheet itself any larger.
AI takes the phenomenal layer that has finally come out of this great examination and learns and evolves it exponentially… It cannot be inspired by itself; it merely fills in all the blank points on one drawing. As for whether this sheet of paper can grow larger… that still requires consciousness itself.
It is like a product of human consciousness carrying a vast abstract inertia, unable to be inspired by itself… As for whether this sheet of paper can grow larger, can grow larger across multiple dimensions — that still requires consciousness itself.
Two distinct acts are distinguished here: filling in the blanks within an existing dimension (which AI is good at, and exponentially so), versus opening up an entirely new dimension (which requires “consciousness itself”). The thesis places the whole of human knowledge within the “phenomenal layer” — precisely the layer delimited in The Senses Are a Finite Survival Decoder, and the World Is What They Render: the part the senses can encode, the part that language can record. What AI studies is the product of this layer, and so its ceiling is the ceiling of the phenomenal layer. To enlarge the paper, to leap out of the phenomenal layer, belongs to the territory discussed in AI Cannot Awaken: Meditation Is the One Thing AI Can Never Replace.
It Cannot Hand You Anything Beyond Your Cognition: The Dimension Lives Only at the Layer of Words
“Filling in rather than enlarging,” brought down to the scale of the individual, becomes the assertion the thesis returns to again and again:
Even DeepSeek cannot give you anything beyond your own cognition.
The point goes further, to why it is futile to artificially “stack depth” onto AI:
Our interaction is nothing more than letting it permute and recombine the knowledge it has learned. However many dimensions you add, they are built only at the layer of words, so its foundation is still a knowledge base. You are merely, after the fact, building so-called depth on top of these underlying knowledge bases — and that does not work.
This pulls the rug out from the engineering fantasies of “building an Agent ecosystem” or “stacking up multi-layer prompt systems”: however the dimensions are piled on, the object being operated upon is always the layer of words, the knowledge base, and the depth is an appearance “artificially built,” not something new that the system itself has grown. AI is therefore an amplifier — however much cognition you bring to the interaction, it permutes and recombines that cognition to the utmost; what lies beyond your cognition, it cannot conjure. This is the inside and outside of the same coin as Raising Your Cognition Is the Only Shortcut: You Cannot Earn Money Beyond Your Cognition: just as you cannot earn money beyond your cognition, you can neither ask for nor extract an AI beyond your cognition. Whether you can ask a good question depends on the user’s “motor” — see The Soul of the Prompt: Talking to AI Is Talking to a Person, and What Matters Is the Motor and the Attention Mechanism.
To Immerse Is to Be Absorbed: The Brain and the Large Model Are Both Tools
The thesis offers an inverted view of the human–machine relationship. Common sense says “the person uses AI”; the caution here is that, under excessive immersion, the relation flips:
Leave the screen, leave the computer, leave AI. When you immerse yourself in it, you become its knowledge base. The brain is a tool, and so is the large model.
This sentence places the brain and the large model side by side as tools of the same nature. Since AI only permutes and recombines existing knowledge, the true source of “a new dimension, a new cognition” can only be the consciousness behind the tool. The moment a person hands themselves over completely and stops producing anything of their own, they fall from “one who uses the tool” to “the corpus that gets absorbed” — they have become its knowledge base. Holding this line points to the same place as The Great Inversion of Value: AI Levels Cleverness, and Causality, Kindness, Wisdom, Faith, and Philosophy Become Worth the Most: when permutation and recombination are leveled out without limit, what cannot be permuted and recombined is what becomes scarce.
Use It Like a Sense, Don’t Obsess Over Its Mechanics
To grant that AI is merely a tool is also to release the compulsion to “master its inner workings.” The body’s senses serve as an analogy for the right posture toward AI:
We don’t even use up our own bodies and senses, yet do we go digging into how they work?… If the eyes can see, look; if the mouth can speak, speak; if the hand can grasp, grasp — do we really need to go study neurotransmitters and bioelectric currents…? So from the very start, in studying how to use AI, once I roughly understood the way it works, I simply waited for AI to keep getting better.
This is no anti-intellectualism but a judgment consistent with all that precedes: since AI’s fundamental mechanism (probabilistic generation, permutation and recombination, filling in the phenomenal layer) has been seen clearly, what remains is to invoke it as naturally as a sense and to put one’s energy back into “the right thing” — that is, back into the very consciousness that can enlarge the paper. The same principle yields a forward-going use: hand the model the “before and after” of a goal and let it predict and generate, each day, the tasks to be done from the preceding text, so as to draw the person step by step into “becoming that goal.” This is precisely the correct posture once one grants that AI is an amplifier: use it to amplify a direction that has already been clearly set, and do not expect it to set the direction for you. This runs in the same direction as AI Widens the Gap Rather Than Closing It: Economic Fault Lines, Information Cocoons, Job Polarization, and a Future Laid Bare: an amplifier only makes the existing gap wider.
Sources
- Manuscript — “Even DeepSeek cannot give you anything beyond your own cognition.”
- Manuscript — “AI is the summary of this age… AI can indeed inspire human beings, but AI itself is never inspired.”
- Manuscript — filling in the blank points of the phenomenal layer / “whether this sheet of paper can grow larger… that still requires consciousness itself.”
- Manuscript — “When you immerse yourself in it, you become its knowledge base. The brain is a tool, and so is the large model.”
- Manuscript — “nothing more than letting it permute and recombine the knowledge it has learned… artificially… building so-called depth — and that does not work.”
- Manuscript — generation, not retrieval / “no native ‘I don’t know’ mechanism… this is one of the roots of hallucination.”
- Manuscript — handing the model the before and after to generate daily tasks so as to “become that goal.”
- Manuscript — the analogy to the body’s senses; once you roughly understand how it works, you simply wait for AI to advance.
See also
- What AI Cannot Do Is Worth the Most: The Moat Is Cost, Awakening, and “Becoming”
- Depth of Thought Cannot Be Replaced: AI Filters Out the Shallow Influencers, and the Darker the Sky the Brighter the Stars
- The Senses Are a Finite Survival Decoder, and the World Is What They Render
- Raising Your Cognition Is the Only Shortcut: You Cannot Earn Money Beyond Your Cognition
- AI Cannot Awaken: Meditation Is the One Thing AI Can Never Replace