Why everyone is talking about Andrej Karpathy's autonomous AI research agent | Fortune

Andrej Karpathy, a prominent AI researcher, recently demonstrated an 'autoresearch' system where an AI agent autonomously conducted 700 experiments to improve the training of a language model, resulting in an 11% speedup. Shopify CEO Tobias Lütke replicated this success with a 19% performance gain on internal data. This experiment sparked discussion about 'recursive self-improvement' in AI and the potential for rapid, uncontrolled advancement. While not fully self-improving, Karpathy believes this approach – which he frames as the 'final boss battle' for AI labs – will become standard practice, accelerating AI research. He envisions collaborative 'agent swarms' exploring optimizations in parallel, applicable to any efficiently evaluable metric. Some critics note similarities to existing AutoML techniques, but Karpathy argues his system, leveraging an LLM to write code and learn from experiments, is significantly more powerful.

(Source：Fortune)