Dwarkesh Podcast

Carl Shulman (Pt 1) - Intelligence Explosion, Primate Evolution, Robot Doublings, & Alignment

75 snips
Jun 14, 2023
Carl Shulman, a research associate at the Future of Humanity Institute at Oxford, dives deep into the fascinating world of AI and its potential. He discusses an intelligence explosion driven by rapid advancements in algorithmic progress and hardware. Shulman explores the implications of primate evolution on AI scaling and how soon AIs could begin conducting their own research. He also shares his thoughts on AI alignment and why he remains optimistic about navigating the risks of a possible AI takeover. Get ready for a mind-bending journey!
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

AI Feedback Loop

  • AI progress accelerates when AI automates research tasks, creating a feedback loop.
  • This automation significantly boosts effective labor, outpacing diminishing returns in hardware improvement.
INSIGHT

Effective Compute Growth

  • AI progress depends on hardware investment, hardware technology, and software progress.
  • Effective compute increases through investment, better models, and cheaper training chips.
INSIGHT

Evolutionary Upper Bound

  • Brute-force evolution created human intelligence, setting an upper bound for AI development.
  • AI benefits from directed development, unlike evolution's random process, skipping early steps.
Get the Snipd Podcast app to discover more snips from this episode
Get the app