Early OpenAI's Tim Shi on how close we are to self-improving AI

agihouse_org · x · 2026-11-21

AGI House publishes a conversation with Tim Shi (early OpenAI, co-founder of Cresta and Recursive) on whether AI can run its own research — propose an idea, test it, decide what's next — and how close we are to self-improving AI. Thread takeaways: Recursive's Eureka Machine automates the research loop, validation (evals, wet labs, trials) is the bottleneck, METR data shows task length doubling every few months, and he advocates know-your-user processes against near-term hacking risks.

Related event: Early OpenAI Member Tim Shi on AI-Driven Autonomous Research(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →