Vibeleaderboard
← All Intel
Intel / video

Rich Sutton and Khurram Javed: Why AI Models Stop Learning, and How to Start It Again

Source
youtube.com
Author
Sequoia Capital
Date

Context

Rich Sutton, a founder of reinforcement learning (training by trial, action and reward) and author of the essay The Bitter Lesson, says in this Sequoia Capital conversation that the term continual learning only exists because the field drifted. For him all learning is continual, since an always acts and learns. He is starting Oak Lab with his former student Khurram Javed.

On language models, Sutton calls them both a positive and a negative example of the bitter lesson, his view that methods which scale with computation win. They scaled by drawing on the internet's text, but that text is finite and small next to the world. He and Javed call this the big world hypothesis: there is always more to learn than any dataset holds, so a system that learns from its own experience could avoid relying on people. Asked about synthetic data, Sutton calls it a big mistake, and the conversation's reasoning is that humans still decide which generated data is good. These are the speakers' positions, not tested results.

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.
  • grounding — Tying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.
Read the source www.youtube.com
More from Sequoia Capital
Recommended reads
Comments

Checking sign-in…

Loading comments…