#self-improvement

An Anthropic researcher just gave us a peek at self-improving AI

This article explains how automated AI systems can self-improve their alignment without sacrificing general capabilities, a breakthrough in AI safety research.