Recursive Self-Improvement

Most technologies improve because people improve them. Recursive self-improvement asks a stranger question: what if the thing being improved is the improver itself?

An AI system that is good at AI research could, in principle, design a slightly better version of itself. That better version is even better at the same task, so it designs a better one still. Each turn of the loop shortens the next.

Why it matters

Everything hinges on how fast the loop turns. If each iteration takes years, you get steady progress you can watch and steer. If it takes days, then hours, a modest lead becomes a sudden chasm. The shape of that curve, slow or fast, is one of the central questions in AI safety.

This page is an early sketch. A full, animated deep-dive is coming.