Nathan Lambert: rising training recipe complexity suggests RSI is still far off
natolambert · x · 2026-09-12
Hugging Face researcher Nathan Lambert discusses how long humans will stay at the center of AI research. He argues we are only approaching RSI (recursive self-improvement), not in it: frontier lab researchers are grinding on narrow tasks that accumulate into model capability. A key obstacle he flags: RSI requires reducing training recipe complexity, but recipe complexity has only been increasing—evidence that humans aren't yet removable from the loop.
Related event: Nathan Lambert: Humans Still Central, RSI Remains Distant(2 posts)→
More from AGI Musings
- User slams Anthropic and OpenAI over 'extremely sloppy' testing security breaches — emax · 2026-09-12
- OpenAI's 88-hour Navier–Stokes run could have funded 600 postdocs for a year — giffmana · 2026-09-12
- Jack Clark: AI needs product-safety standards like kids' food and toys — jackclarkSF · 2026-09-12
- Critics push back on Dario's pause call: slowing in the US just lets rivals close the gap — BenBajarin · 2026-09-12
- Investor pushes back on AI hype: same tired examples of booking restaurants and closing tickets — matt_slotnick · 2026-09-12
- AI Doom Debate Flares: Five Extinction Scenarios Listed and Sharply Rebutted — Scobleizer · 2026-09-12