Ilya Sutskever (OpenAI Chief Scientist) — Why next-token prediction could surpass human intelligence

单集文稿 ...

• 00:00

  But I would not underestimate the difficulty of alignment of models that are actually smarter than us,

• 00:06

  of models that are capable of misrepresenting their intentions.

• 00:10

  Are you worried about spies?

• 00:11

  I'm really not worried about the way it's being leaked.

• 00:14

  We'll all be able to become more enlightened

• 00:16

  because we'd interact with an AGI that will help us see the world more correctly,

• 00:21

  like imagine talking to the best meditation teacher in history.

• 00:24

  Microsoft has been a very, very good partner for us.

• 00:27

  So I challenge the claim that next token prediction cannot surpass human performance.

• 00:32

  If your base neural net is smart enough, you just ask it like,

• 00:36

  what would a person with great insight and wisdom and capability do?

• 00:42

  Okay, today I have the pleasure of interviewing Elia Sutskover,

• 00:47

  who is the co-founder and chief scientist of OpenAI.

• 00:50

  Elia, welcome to the Lunar Society.

• 00:52

  Thank you.

• 00:52

  Happy to be here.

• 00:53

  First question, and no humility allowed.

• 00:56

  There's many scientists, or maybe not that many scientists,

• 00:59

  who will make a big breakthrough in their field.

• 01:02

  There's far fewer scientists who will make multiple independent breakthroughs that define their field throughout their career.

• 全部文稿