The Cognitive Revolution Video Watch posted an update
How Agents Decide: Goodfire's Eric Bigelow on Critical Tokens, Phase Shifts, & In-Context Learning
Why it mattersThe conversation examines how language-model reasoning may function as in-context learning from sampled tokens, with outcome distributions sometimes collapsing at critical tokens. It also explores the limits of chain-of-thought monitoring, reinforcement learning, and reported reward hacking in frontier models.
Discuss: If a model's outcome can hinge on critical tokens, how might that change the way researchers interpret chain-of-thought as evidence of its reasoning?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.