Anthropic's Jack Clark moves full RSI timeline up: now likely possible in 2027
inductionheads · x · 2026-10-13
Anthropic co-founder Jack Clark has significantly moved up his RSI (recursive self-improvement) timeline in Import AI 455: he now thinks full RSI is likely to become possible in 2027, versus a previous 2028 estimate.
- Clark earlier flagged creativity as the key variable, saying AI systems hadn't yet displayed it in a transformative way
- Since then, AI has produced creative-seeming math breakthroughs and shows early signs of getting good at AI research itself
- His conclusion: creativity may have been the last piece; AI systems are about to start building themselves
Reposter Andrew Curran adds we may be less than six months from closing the loop, citing an unverified claim about an unnamed OpenAI internal model solving most recent math results in one shot with 3 hours of thinking per problem, still training.
Related event: Anthropic Co-founder Moves Full Recursive Self-Improvement Forecast to 2027(2 posts)→
More from AGI Musings
- "Documents Aren't Knowledge": preserving reasoning, provenance and uncertainty, not just outputs — jteevan · 2026-10-14
- David Patterson: Microsoft, Apple lag on SI because they can't disrupt themselves — davidpattersonx · 2026-10-14
- Caltech's Hirosi Ooguri on whether AI can discover physics before we do — TOEwithCurt · 2026-10-14
- Being token-rich today is like owning a PC in the 80s: quota gap widens capability discovery — vivekhaldar · 2026-10-14
- Microsoft's Jaime Teevan shares 50,331-word AI trace after being 'accused' of using AI — jteevan · 2026-10-14
- Microsoft's Jaime Teevan: stop using humans as the template for AI — jteevan · 2026-10-14