Articles on Articles — akciali.com

Gödel's incompleteness: the truths no machine can reach

Mon, 22 Jun 2026 09:00:00 +0200

The whole theorem in one picture: what a machine can prove is a strict subset of what is true. The sentence G lives in the crescent that overflows — true, yet forever out of reach.

Imagine you could build a truth machine. You hand it some starting axioms and a few rules of deduction, you switch it on, and it prints out, one by one, every mathematical truth — never wrong, never forgetting one. This is not a science-fiction fantasy. It was, very nearly word for word, the program that one of the greatest mathematicians in history, David Hilbert, set for his entire discipline at the start of the twentieth century. His dream was to put mathematics on fully mechanical rails: a system in which “being true” and “being provable” would be one and the same thing. A mind could then be replaced, for the purpose of doing mathematics, by a blind procedure.

From hands to intention: how an interface creeps closer to the thought

Sun, 21 Jun 2026 09:00:00 +0200

The lesson hiding in a fist

A few weeks ago I gave a 3D memory map a webcam and a pair of hands — point to light up a thought, make a fist to grab the cloud, spread two fists to zoom. The write-up was mostly a story about failure: the obvious gesture (pinch-and-drag, like a touchscreen) was miserable, and the fix was to make the fist — the posture the hand keeps falling into anyway — the verb.

A 30B model on an 8 GB GPU: a small win with Mixture-of-Experts

Sat, 20 Jun 2026 09:00:00 +0200

A small, satisfying win

A while back I moved my small home AI stack off Ollama and onto raw llama.cpp, running on a single 8 GB GPU — an RTX 2080, a card old enough to vote in dog years. The chat side was an 8-billion-parameter model that fit comfortably on the card; an embedding model ran on the CPU. It worked well, and I left it alone.

But I wanted more: better reasoning, and noticeably better French, for the one job that model does — writing a short daily status digest. The obvious path was a bigger model, and the obvious objection was that I had no bigger GPU to put it on. This is the story of getting a 30-billion-parameter model to run on that same 8 GB card anyway. It is a small thing. It was also very satisfying.

The free energy principle: why staying alive means refusing surprise

Wed, 17 Jun 2026 09:00:00 +0200

One gap, two remedies. Perception bends the model onto the world; action bends the world onto the model. Both close the same gap.

There is a theory in contemporary neuroscience that is at once the most discussed and the most contested of its kind. It is the free energy principle, forged by the neuroscientist Karl Friston, and its ambition is frankly outsized: to explain, from a single imperative, why a living thing perceives, acts, learns, and even why it exists as a “thing” set apart from the rest of the universe. The claim is large enough to invite suspicion, and the suspicion is healthy. But the idea is worth climbing, and it is best climbed by images.

The feedback loop: the science that blurred the line between machine and living thing

Sat, 13 Jun 2026 09:00:00 +0200

The universal loop: act, measure the result, correct the gap. The same diagram describes a thermostat and a living body.

In 1948 a mathematician at MIT, Norbert Wiener, published a book whose title he had just invented: Cybernetics, from the Greek kubernētēs, “the steersman,” the one who holds the tiller. The idea behind it sounds banal and is not banal at all: a system can govern its future behaviour using its own past performance. It acts, it measures the result of that action, and it uses the measurement to correct the next action. That is the whole of it. The arrangement is called a feedback loop, and Wiener made a claim that was, for his time, faintly scandalous: this same loop describes a missile chasing an aircraft, a thermostat holding a room at twenty degrees, a hand closing around a glass, and a body keeping its blood sugar in range. The same pattern, whether the parts are metal or flesh. That is the moment when, intellectually, the boundary between the machine and the living thing began to dissolve.

Becoming lucid in a dream: the science of oneironautics

Mon, 08 Jun 2026 09:00:00 +0200

The lucid dreamer signals from inside the dream with deliberate eye movements — the first letters ever sent from the far side of sleep.

Here is a sentence that ought not to be true: there is a state in which one knows one is dreaming, while dreaming, and can act on the dream without waking up. Not a blurry half-waking, not a memory reconstructed in the morning — a full consciousness lodged inside the film, able to say “this is a dream” and to decide what happens next. This is the lucid dream, and it has the rare distinction of being both an ancient practice (Tibetan Buddhists cultivated it as a “dream yoga”) and a rigorously demonstrated laboratory phenomenon. It is worth seeing how it was proven, what is happening inside the skull, and above all how it is caught.

Reaching into the cortex: steering a memory map with your bare hands

Wed, 03 Jun 2026 08:30:00 +0200

Wave at your own memory

A while back I turned an AI assistant’s memory into a 3D star map you can fly through — every glowing point a note, distance standing in for meaning, colour for age. It was a mouse-and-keyboard thing: scroll to zoom, drag to spin.

This is the sequel. I put a webcam behind it and asked a simpler, sillier question: what if you could just reach in with your hands?

Inside the Mesh: a live map of a self-hosted lab

Mon, 01 Jun 2026 09:00:00 +0200

Start here: a lab you can watch breathe

Open the demo — mesh-demo.akciali.com — and you are looking at a neon graph drifting on near-black. The bright node in the middle is the machine that runs everything. The smaller glowing nodes around it are the services and agents that make up my home lab. Lines show who talks to whom, colour shows health, and a console scrolls down the right-hand side with real events as they happen. It is, quite literally, a control room for one person’s lab.

From Ollama to llama.cpp on a single 8 GB GPU

Fri, 29 May 2026 09:00:00 +0200

A field report on moving a small self-hosted AI stack off Ollama and onto raw llama.cpp — what went well, and the eight things that bit me on the way.

Update — June 2026: the 8B chat model described below has since been replaced by a 30-billion-parameter Mixture-of-Experts model running on the same 8 GB card, via CPU offload of the idle experts. See the sequel: A 30B model on an 8 GB GPU.

What makes you 'you'? Personal identity and the problem of continuity

Sun, 24 May 2026 09:00:00 +0200

Identity through time looks less like an object that persists and more like a chain of overlapping states held together by thin threads.

Start with a fact that ought to be more disturbing than it feels: a person is almost nothing of what they once were. The overwhelming majority of one’s atoms have been replaced several times since birth. Opinions drift, tastes turn over, memories fade and reshape themselves. The reader finishing this sentence shares not a single skin cell with the child they once were. And yet we all say, without a flicker of doubt, “that was me.” On what, exactly, does this “me” rest, when the matter that supposedly carries it does not survive at all? This is the problem of personal identity, and it is far more slippery than it looks.

How a gesture becomes a reflex: motor learning and the ceiling that isn't there

Mon, 18 May 2026 09:00:00 +0200

A jagged trajectory smoothing into a clean curve: learning a movement is, quite literally, rewiring the electrical plumbing of the nervous system.

Start with the fact that turns the intuition inside out: an expert movement is not a thought performed faster — it is a movement that has left thought entirely. A beginner’s brain steers every step consciously: the stance, the grip, the alignment, the breathing, the release. It is slow, costly, fragile under stress. The expert is not doing “the same thing, better.” The expert is doing something else: the movement executes on its own, beneath the radar of awareness, while attention is freed for the things that actually decide the outcome — the wind, the rhythm, the chosen instant. Learning a gesture is not learning to perform it better. It is learning to perform it without having to think about it. Everything else follows from there.

Inside the Cortex: turning an AI's memory into a 3D star map

Tue, 12 May 2026 09:00:00 +0200

Start here: a galaxy made of thoughts

Open the demo — cortex-demo.akciali.com — and you are looking at a cloud of glowing points drifting in 3D. You can grab it, spin it, fly into it. Each point is a single memory: one note, one fact, one decision. Thin threads run between some of them — those are links, thoughts that reference one another. A few points pulse brighter than the rest: those are the recent memories, the things learned in the last day.

Wiring a brain to a machine: how a thought becomes a command

Tue, 05 May 2026 09:00:00 +0200

A brain-machine interface taps the motor output line at the source and reroutes it — to a cursor, a synthetic voice, a robotic arm.

Start with the fact that tips the whole intuition over: a fully paralysed person can move a cursor, type, and even “speak” — without moving a single muscle. Not by magic, but because paralysis breaks the cable between cortex and muscles while leaving the cortex intact. When the person forms the intention to move a hand, the motor cortex lights up exactly as it did before. The command signal is still being emitted; it is simply no longer delivered. A brain-machine interface (BMI, or BCI) does one thing, and that one thing is vertiginous: it goes and fetches the signal at the source and plugs it in somewhere else — into a cursor, a speech synthesiser, a robotic arm.

The factory of dreams: what your brain builds when no one is watching

Sun, 26 Apr 2026 09:00:00 +0200

Sleep is not a uniform plunge into the dark. It is a generator that keeps running with the sensors switched off.

Start with the fact that quietly rearranges everything: almost everyone dreams for roughly two hours a night. Not just “people who remember their dreams” — everyone, every night. When a sleeper is woken during REM sleep, they report a dream in progress around eight times out of ten, whether or not they consider themselves a dreamer. The difference between someone who remembers and someone who does not is not a difference in production. It is a difference in recording. The problem was never the dreaming. The problem is that the tape does not get written to disk.

The attention mechanism: how a machine learned to choose what to look at

Sun, 19 Apr 2026 09:00:00 +0200

Attention: every word queries the whole sentence and feeds mostly on the words that answer best.

Read this sentence: “The glass fell off the table, but it did not break.” The word it — what does it refer to? The glass, obviously. Not the table. You knew it instantly, effortlessly. But pause on what just happened. In a twelve-word sentence, your mind knew that to understand “it” you had to go and look at “glass” — and ignore “table,” “fell,” “but.” It distributed its attention. That is precisely the problem a machine must solve to understand language, and the solution is called, fittingly, the attention mechanism. It is the single idea that flipped the whole field of AI in 2017 and is, literally, the engine inside every large language model in use today.

Augmenting human intellect: the idea Engelbart had 60 years too early

Sun, 12 Apr 2026 09:00:00 +0200

Augmentation: not a machine that replaces the human, but a lever that raises what the human-plus-tool can do.

December 9th, 1968, San Francisco. A conference hall, a thousand engineers, the lights dimmed. On stage, a calm 43-year-old with a headset and an odd wooden box under his right hand. For ninety minutes, almost without raising his voice, Douglas Engelbart walks a stunned audience through: the mouse, on-screen windows, hypertext (clickable words that jump elsewhere), real-time text editing, and — the climax — two people at a distance editing the same document while seeing each other over video link. In 1968. While the rest of the computing world was still punching holes in cardboard cards.

The Chinese Room: is understanding the same as computing?

Sat, 04 Apr 2026 09:00:00 +0200

The Chinese Room: perfect symbol-shuffling on the outside, nobody home on the inside.

There is a tempting modern thesis: to understand something is to compress it. A model that predicts the next token is, in effect, searching for the short program that explains its training data, and the better it compresses, the better it seems to “understand.” The relationship can even be measured. And just as that idea starts to feel settled, a voice from 1980 cuts in: not so fast — manipulating symbols according to rules, however well, is still not understanding. That voice belongs to John Searle, and his weapon is the Chinese Room. It is the most argued-over thought experiment in all of philosophy of mind, and the rise of large language models has dragged it back to the centre of the table.

Kolmogorov complexity: the true amount of information in an object

Sat, 28 Mar 2026 09:00:00 +0200

The information in an object is the size of the shortest recipe that recreates it — not its apparent length.

Shannon’s entropy measures the average surprise of a source, computed over a distribution of probabilities. But there is a question it cannot quite reach: a single object — this file, that sequence, with no source and no probabilities behind it — how much information does it really contain? The answer, given by a twenty-five-year-old Russian mathematician in 1965, is at once the most beautiful definition of information ever written and one of the most unsettling, because it leads straight to the impossible.

The predictive brain: an organ that spends its life minimising surprise

Sat, 21 Mar 2026 09:00:00 +0200

Predictions flow down the hierarchy; only the error — the part the model failed to anticipate — flows back up.

Information theory gives us a precise notion of surprise: an event of probability p carries −log₂ p bits, so a rare event carries a great deal of information. It is striking to take that same quantity and place it at the heart of a 1.4-kilogram organ that spends its entire existence fighting it. Under one increasingly influential reading, the brain is, at bottom, a machine for reducing surprise.

Skills as an orientation map: how an agent loads expertise on demand

Tue, 17 Mar 2026 09:00:00 +0200

A general agent, and the wall of instructions

A capable general-purpose agent will do most things passably. Getting it to do your specialised work, your way — the house report format, the domain checklist, the procedure with the three non-obvious steps — is a different problem. The naive fix is to paste all of that into the prompt. It fails twice. Context is finite, so the procedures crowd out the actual task; and standing instructions are always on, competing for the model’s attention even when they are irrelevant to what you are doing right now. Past a certain size, a wall of instructions makes the agent worse at everything.

Entropy: the measure of our ignorance

Sat, 14 Mar 2026 09:00:00 +0200

Entropy counts the configurations we cannot tell apart — the gap between what we measure and what we ignore.

Ask ten people what entropy is and nine will answer “disorder.” It is the popular image: the room that never tidies itself, the coffee that cools, the universe drifting toward chaos. The picture is not wrong, but it hides the essential point. Entropy is not a property of objects. It is a property of our knowledge of objects — a measure of uncertainty. Once that shift is made, three disciplines that seemed unrelated turn out to be the same thing seen from three angles.