John Langford on post-RSI research: asking interesting questions comes last
JohnCLangford · x · 2026-11-15
ML theorist John Langford and MIT researcher Dimitris Papail discuss what research looks like in a post-RSI (recursive self-improvement) era. Langford expects the last human-held part of research is asking the interesting questions. Papail counters that RSI may automate 'high-taste question asking' too. Langford says this may happen, but RSI is a spectrum, not a switch: research multipliers grow gradually as more of 'taste' gets automated — a 10x research speedup seems reasonable today, another order of magnitude within a year unsurprising, and 'full RSI' may not change the shape of the future.
More from AGI Musings
- "100,000 people make AI more powerful, only 300 make it safer" — claim sparks dispute — jd_pressman · 2026-11-16
- Transhumanism is 'dystopia dressed as medicine', argues viral critique of Silicon Valley ideology — DionysianAgent · 2026-11-16
- 'Commanding Heights' paper: AI drives massive change as capitalism's tool, no singularity — code_star · 2026-11-16
- Bill Gates: AI is more dangerous than nuclear weapons but could benefit more than nuclear energy — VraserX · 2026-11-16
- Blogger: the ultimate breakthrough is manufacturing ASI until intelligence is no longer scarce — Dr_Singularity · 2026-11-16
- Robin Hanson: Doomsday theories must predict smaller testable events — QuintinPope5 · 2026-11-16