Vibeleaderboard
← All Intel
Intel / video

OpenAI’s Deep Research Team on Why Reinforcement Learning is the Future for AI Agents

Source
youtube.com
Author
Sequoia Capital
Date

Context

Deep research is an OpenAI in ChatGPT that searches many websites and writes a long report, taking roughly 5 to 30 minutes. In this Sequoia Capital interview with team leads Isa Fulford and Josh Tobin, the host says it was trained end to end with reinforcement learning (the model practices tasks and is rewarded for good outcomes) on hard browsing and reasoning tasks, and a team member says the methods resemble those used for reasoning models.

An unidentified speaker in the interview's opening clip argues that you get what you optimize for. Gluing together models not optimized for the final task tends to do worse than optimizing directly for it, so reinforcement learning on top of models is expected to be critical for the most powerful agents. That is a design view from the OpenAI side, and no comparison data accompanies it.

Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Read the source www.youtube.com
More from Sequoia Capital
Recommended reads
Comments

Checking sign-in…

Loading comments…