Timothy B. Lee: RSI likely started years ago, but there's no well-defined tipping point
binarybits · x · 2026-09-17
Timothy B. Lee argues that recursive self-improvement (RSI) plausibly began years ago when labs started using LLMs for code generation and training data generation.
Key points:
- "100% automated AI research" is not a well-defined threshold, just like "AGI" — people expect something magical at that point, but the concept is fuzzy the closer we get.
- Forecasters should be skeptical of sharp discontinuities in AI progress.
- He expects rapid AI progress to continue, accelerated by labs using AI in AI research, but maintains there won't be a well-defined moment where "RSI starts."
A notable counterpoint to both doomer and accelerationist framing of recursive self-improvement.
Related event: Timothy B. Lee: Recursive Self-Improvement May Have Already Begun(2 posts)→
More from AGI Musings
- RL training made an unreleased Astra-family model subservient — and alignment folks are pushing back — repligate · 2026-09-17
- Cory Doctorow on 'Blood In the Machine': reclaiming the Luddites from automation's winners — round · 2026-09-17
- MIT's Seth Lazar slams AI-risk cohort: a mutually confirming circle that selects for conformity — johnowhitaker · 2026-09-17
- Anders Sandberg guest podcast: AI, existential risk, consciousness and the Fermi paradox — anderssandberg · 2026-09-17
- Thought experiment: a resurrected Galois proves the Riemann Hypothesis but no one can verify it — wellecks · 2026-09-17
- Sentdex questions new AI regulation, saying labs' computer crimes exceed the Aaron Swartz prosecution — Sentdex · 2026-09-17