What is the research speed multiplier of the most advanced current LLMs?

post by wunan · 2024-02-21T12:39:11.034Z · LW · GW · No comments

This is a question post.

Contents

  Answers
    5 Tomás B.
    3 Nathan Helm-Burger
None
No comments

In Sam Altman's recent interview on Bill Gates' podcast, he said AI tools currently can speed up a programmer by about 3x. It's unclear whether he's referring to publicly available tools like copilot, or internal tools at OpenAI.

  1. Does anyone know whether he's referring to private or public tools? Is that 3x figure based on any public research?
  2. More speculatively, if programmers are sped up by 3x, how much are AI researchers sped up, particularly using whatever internal tools are currently available at OpenAI? The job involves more than just programming, but many non-programming tasks can be sped up as well.
  3. Even more speculatively, how much do you predict GPT-5 will speed up researchers at OpenAI?

Answers

answer by Tomás B. · 2024-02-21T15:23:17.427Z · LW(p) · GW(p)

GPT-5 with a context window that can fit entire code bases is going to be very scary. Particularly if you think, as I do, that agency is going to start to work soon. I really do think at least "weak recursive self improvement" of the form of automating AI research/training loops is on the table relatively soon. 

No comments

Comments sorted by top scores.