I did appreciate the discussion on the actual cost. I've been hearing a lot of that $5M number being thrown around, and while it definitely felt unlikely, I didn't have the context. In particular [1] linked from the post was really helpful.
Framing and Apple fanboying as it is. It just takes time to integrate, deploy and benchmark new models, that's the whole secret. The first big confirmation that DeepSeek-R1 is good was published only few days ago[1], it just takes some time to crowdsource votes.
I did appreciate the discussion on the actual cost. I've been hearing a lot of that $5M number being thrown around, and while it definitely felt unlikely, I didn't have the context. In particular [1] linked from the post was really helpful.
[1] https://www.interconnects.ai/p/deepseek-v3-and-the-actual-co...
thanks for this. insightful analysis on deepseek’s costs.
Framing and Apple fanboying as it is. It just takes time to integrate, deploy and benchmark new models, that's the whole secret. The first big confirmation that DeepSeek-R1 is good was published only few days ago[1], it just takes some time to crowdsource votes.
[1] https://x.com/lmarena_ai/status/1882875989610594542