LessWrong (Curated & Popular)

"Two-year update on my personal AI timelines" by Ajeya Cotra

6 snips
Sep 22, 2022
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Timelines Shifted Earlier And Narrower

  • Ajeya Cotra shifted her TAI timeline notably earlier: ~15% by 2030, ~35% by 2036, median ~2040.
  • This concentrates probability mass into a narrower 2026–2050 window compared to her prior 2032–2064 spread.
INSIGHT

Automating AI Research Lowers The Bar

  • Cotra now treats automating AI research/engineering as a lower bar for transformative impact than automating all science.
  • She reasons that automating AI R&D can create fast feedback loops that rapidly accelerate AI progress itself.
INSIGHT

Short Horizon Training Seems More Plausible

  • She became more confident that short effective training horizons can produce transformative capabilities without extensive meta-learning.
  • This reduces required model size and effective horizon for TAI compared to her earlier view.
Get the Snipd Podcast app to discover more snips from this episode
Get the app