I haven’t gotten to the finance part yet but his discussion on the state of AI currently is excellent. I’m going to post something later about my experiments with DS-R1 in offline form and playing with “the big model” through their apps. The differences are pretty interesting and the big model is quite impressive.

Are you making any investment decisions today? I was thinking it might be a good day/week to “buy the dip” assuming NVidia actually is still a valuable company and will likely experience some overcorrection in the opposite direction. It’s been jittering all day but the downward trend continues for the moment.

Edit:

Wanted to add that I also read this today and he has some different insights to DS as well. In particular, he thinks DS is being honest in their costs, and does some analysis on how their optimizations make their claims of computing on a lot of H800 chips plausible.

He also talks about how DS-R1 uses a somewhat different chain-of-thought/reasoning model than ChatGPT o1. They used RL (no HF!) to have DS “teach itself” how to reason. To me, that’s maybe the most alarming bit. I’d been confident this approach wouldn’t get to AGI but with these techniques… I dunno, maybe we’re much closer to bootstrapping and the AGI revolution than expected? Like…within 2 or 3 years maybe?

Ah, the OP also mentions this:

With R1, DeepSeek essentially cracked one of the holy grails of AI: getting models to reason step-by-step without relying on massive supervised datasets. Their DeepSeek-R1-Zero experiment showed something remarkable: using pure reinforcement learning with carefully crafted reward functions, they managed to get models to develop sophisticated reasoning capabilities completely autonomously. This wasn’t just about solving problems— the model organically learned to generate long chains of thought, self-verify its work, and allocate more computation time to harder problems.