Machine Learning Street Talk (MLST)

Can We Develop Truly Beneficial AI? George Hotz and Connor Leahy

8 snips
Aug 4, 2023
George Hotz is a renowned tech innovator known for jailbreaking devices, and he's currently focusing on AI at MicroGrad. Connor Leahy emphasizes AI safety at Conjecture. They debate whether creating beneficial AI aligned with human values is achievable, with Hotz arguing against it while Leahy believes it's solvable. The discussion also touches on the distribution of AI power to prevent dominance and the urgent need for alignment to avoid danger. They tackle the balance between open access and governance in AI's future, concluding that coordination is essential.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Inevitable Intelligence Growth

  • The increase in global power and intelligence is inevitable, barring a major catastrophe.
  • Distributing AI widely prevents single-entity domination, so open-sourcing is crucial.
INSIGHT

Misuse and Control Problems

  • Misuse of controlled AGI by bad actors poses a severe risk, potentially leading to suffering worse than death (S-risks).
  • Unsolved technical control problems may lead to AIs fighting amongst themselves, ignoring humans entirely.
ANECDOTE

Unexpected Civility

  • Connor Leahy saw a video of a parliamentarian being accosted, but nobody resorted to violence.
  • This surprising civility demonstrates unusual coordination and alignment within modern society.
Get the Snipd Podcast app to discover more snips from this episode
Get the app