Anthropic's Jack Clark moves up RSI timeline: full recursive self-improvement likelier in 2027 than 2028
haider1 · x · 2026-10-13
Anthropic co-founder Jack Clark has revised his recursive self-improvement (RSI) timeline: he now thinks full RSI is likelier to become possible in 2027 than 2028.
- Rationale: AI systems have shown tremendous creative-seeming math breakthroughs recently, and benchmarks like TasteVal (P-Zero Research) suggest they're getting better at AI research itself.
- He had earlier argued creativity was the key variable for self-improvement speed, noting at the start of the year that AI hadn't yet displayed it in a transformative way.
Replier Haider adds OpenAI/Anthropic could hit RSI as soon as 2027, but cautions that self-improvement still faces diminishing returns — every iteration requires compute — and alignment remains unsolved, so no automatic rapid intelligence explosion.
Related event: Anthropic Co-founder Moves Full Recursive Self-Improvement Forecast to 2027(2 posts)→
More from AGI Musings
- Investor: personal AI agents will become the front door to healthcare — mattlungrenMD · 2026-10-13
- For 300 years a computer was a person: what that history says about which skills AI erases — r0ck3t23 · 2026-10-13
- Anthropic's Jack Clark moves full RSI timeline up: now likely possible in 2027 — inductionheads · 2026-10-13
- Economist Paul Novosad: AI-boosted workers often can't tell their output is bad — paulnovosad · 2026-10-13
- Startup advice: use the best AI first, optimize cost later — evisoft · 2026-10-13
- Anthropic's constitution called best record of its alignment tradeoffs as Claude trained to balance principals' interests — austinc3301 · 2026-10-13