Allison @adjectiveallison.com · Nov 26 Aidan Mclau over on x wrote some great content that you shouldn't miss out on just because you don't go to the site. It's a fantastic article on reasoning models. Give it a read. aidanmclaughlin.notion.site/reasoners-pr...
7 likes 2 replies
?
Replies
The RL piece seems sort of tacked on, and I’m not surprised it nerfs the models in other tasks. It seems like you could get “reasoning-style” models just via prompting to do CoT (and I guess that’s what Claude sonnet does)
i read it earlier today, i'm not sure i buy the premise. i think trying to build larger models to one-shot hard problems is the wrong way to AGI. i think the labs agree, which is why they're trying to build AI models that solve tough problems over a long-time horizon.