Recursive self-improvement through collective intelligence

The article discusses the concept of recursive self-improvement (RSI) in AI models, where the model itself acts as its own optimizer and evaluator. The authors introduce Multi-Agent Self-Supervision (MASS), a learning loop that uses the same model for executor, optimizer, and evaluator roles. By distilling collective intelligence from multi-agent systems, MASS aims to improve the model's performance on non-verifiable research tasks, where automatic rule-based signals are not available.

The process is evaluated using GPT-5.5 and Claude Opus 4.8, which provide an outside measure of improvement in a loop where supervision comes solely from the model.

Brief written by urgent.news from Hacker News's own syndicated text. Machine-written — may contain errors; check the original before relying on it.