An aircraft passing overhead led me to a strange thought about distance. Someone can be physically close yet inhabit a completely different world, while a point only ten feet from an airline passenger can be effectively unreachable without life-supporting technology. That observation suggests a broader way of looking at reality: as countless overlapping conceptual landscapes containing gradients of value, knowledge, trust, accessibility, scarcity, risk and possibility. Some of those gradients are unusually steep, where a small change in state produces a disproportionate change in outcome. Those places often resemble what we call opportunity. But steepness alone is not enough. The more useful challenge is finding the consequential gradients: the ones that actually govern outcomes, can be traversed economically, and perhaps have only recently become accessible. AI may not only change those gradients; it may eventually help us discover ones we have never thought to look for.
Writing
Essays, papers, notes, and references.
A place for longer pieces, formal work, and useful written records.
An aircraft passing overhead led me to a strange thought about distance. Someone can be physically close yet inhabit a completely different world, while a point only ten feet from an airline passenger can be effectively unreachable without life-supporting technology. That observation suggests a broader way of looking at reality: as countless overlapping conceptual landscapes containing gradients of value, knowledge, trust, accessibility, scarcity, risk and possibility. Some of those gradients are unusually steep, where a small change in state produces a disproportionate change in outcome. Those places often resemble what we call opportunity. But steepness alone is not enough. The more useful challenge is finding the consequential gradients: the ones that actually govern outcomes, can be traversed economically, and perhaps have only recently become accessible. AI may not only change those gradients; it may eventually help us discover ones we have never thought to look for.
An experiment with E. coli produced an intriguing evolutionary result. Forced to balance the competing demands of growth and chemotaxis, populations adapted until they reached what appeared to be an adaptive peak, where further improvement seemed constrained by a hard energetic trade-off. Then some escaped. They had not abolished the trade-off, but had evolved a way to express different behaviours at different times: grow when growth mattered, move when movement mattered. The result offers a useful way to think about modern AI. GPTs are extraordinarily effective products of optimisation within vast parameter spaces, yet increasingly their capabilities depend on systems built around them: retrieval, memory, tools, variable inference-time computation and orchestration. Sometimes a stubborn frontier really is a limit. Sometimes it reflects the dimensions in which we chose to describe the problem.
We routinely discard data as noise because it is not relevant to the measurement we intended to make. But irrelevance is not the same as absence of information. Physical events leave traces across sound, vibration, temperature, electrical current, radio emissions and other signals, often simultaneously. Cheap sensors can now capture many of these channels at once, while self-supervised machine learning can discover relationships between them without requiring every useful property to be labelled in advance. This article explores a shift from building sensors to answer predetermined questions towards collecting rich physical signals and learning what they can tell us. The central question is simple: what does the noise already know?
I went back to the simplest equation in school algebra and found a neuron inside it. This is what happened when I followed that thread as far as it would go, through language models, biological minds, and the evolved interface we mistake for reality, and arrived somewhere I didn’t expect: a shoreline, not a wall.
We've convinced ourselves that clarity is something you can acquire, read the right list, name the right bias, and step clear of the mess everyone else is stuck in. This is about why that doesn't work, and why the failure is worse than it looks. The frameworks that explain how your judgement fails are mostly correct, and that's precisely the problem: a false map you escape by noticing it's false, but a true one you move into and furnish. Detachment doesn't help, the certainty that you're the exception is itself one of the oldest seats in the room. What's left isn't another framework to collect but a single unglamorous habit: turning, deliberately, toward the evidence that costs you something. It won't make you immune. It just shifts the odds, late and unreliably, for anyone willing to stop wanting it cheap.
The popular fear of AI is trapped inside the wrong film. The real threat is not Terminator rupture but Matrix replacement: not that machines kill us, but that they break the scarcity mechanisms that give people status, and once status breaks, reality itself becomes negotiable. AI commoditises the middle layer of human competence, hollowing out the professional class's claim to be necessary. Capitalism answers abundance with enclosure; value migrates from things to scarce worlds; and people retreat into personalised synthetic realities warmer than the world outside. The end need not arrive as apocalypse. It can arrive as comfort, as a bespoke interface, as abundance so complete that human beings forget which scarcities made them real.
“Make money while you sleep” is an old dream, and AI has made it newly plausible: a machine can research, draft, sort, and prepare while you rest. But the phrase confuses production with revenue. A draft is not a sale; output overnight is not income. AI is neither employee, oracle, nor entrepreneur. It is a layer of executable cognition wrapped around repeatable work, and in the hands of a fantasist it produces fantasy at scale, in the hands of an operator, leverage. The machine does not make money while you sleep. It performs the labour that used to stop you building the thing that makes money, clearing the path between judgement and output. The old world rewarded those who worked hardest for longest. The next rewards those who can tell labour from judgement, and automate only the first.
A pump can look steady on every gauge while its controller quietly works harder underneath, the surface reading and the real condition aren't the same thing, and most sensing systems can't tell the difference until it's too late. This essay makes the case for training a model to forecast a hidden representation of what a system will do next, rather than the raw numbers themselves, then teaching a separate stage to translate that representation back into something readable. It works through the architecture from first principles, encoder, predictor, decoder, the role of context from other sensors and written records, before following the idea into three places it might matter: catching faults before they become failures, spotting the moment physical and digital behaviour stop agreeing with each other, and finding ways to let a person follow a system's condition by ear instead of by staring at a dashboard.
An under-16 social media ban need not become a national identity layer for the internet. The fear that age checks are digital ID through the back door is well founded, but only if we ask the wrong question. The right one is not who this person is, but whether a platform can know only that they are over 16. Privacy-preserving credentials and zero-knowledge proofs answer it: a narrow, unlinkable, session-bound proof that reveals no name, no face, no document. The standard should be brutal and simple: prove the attribute, not the person.
Every argument about where AI goes next is, underneath, an argument about the climb: recursion, self-improvement, compute bending the curve back on itself and accelerating. The progress is real and the excitement is earned. But a climber is only as good as the hill. Optimisation power, however vast, is worthless without something faithful to climb toward, and that target, the gradient that tells the system which way is up, is the thing nobody is pricing. We have built an extraordinary engine for going up, and said almost nothing about who decides where up is. This piece argues that the unpriced variable in the whole debate is not capability but direction.