About this survey

The Jagged Frontier began as a single long conversation about a deceptively simple question: how much smarter will these systems actually get, and at what? The conversation wandered — from why a model miscounts the letters in strawberry, to whether anything could write a new Faust, to AI-generated pop songs, to flight-control software, to whether a machine could ever solve a problem of Erdős. This site is that conversation rebuilt as a survey: five chapters that take the limits of AI one terrain at a time and look for the structure they share.

The voice

The chapters are written in the voice of an analytical observer — calm and explanatory rather than ideological, concrete before abstract, and careful to keep observation, interpretation, and speculation apart. The aim is neither hype nor alarm. Where a foundational idea is being reopened by new technology — Hayek's knowledge problem by large-scale machine learning, for instance — the survey tries to give the strongest current challenge and the strongest rejoinder, and to land where the evidence actually is.

The method

Each chapter was drafted, then handed to an adversarial editor for critique — cuts, clarity flags, comprehension checks — and revised against that critique before publication. The intent of that loop is the one stated in the survey itself: when generation is cheap, the value moves to judgment and verification. It seemed only honest to build the site the way its argument recommends.

What it is not

This is not a forecast with dates, and not a settled verdict. The recurring claim is structural rather than chronological: the frontier is a coastline, not a wall being crossed on a particular day. The figures cited in each chapter are drawn from the source conversation and are best read as the state of a fast-moving field at the time of writing, not as fixed coordinates.