We don't actually know how much money DeepSeek spent or how much compute they used. The numbers being thrown around are suspect, the paper they published didn't reveal the costs of all models nor the R&D cost it took to develop them.
In any AI R&D operation the bulk of the compute goes on doing experiments, not on the final training run for whatever models they choose to make available.
One thing I (intuitively) don't doubt - that they spent less money for developing R1 than OpenAI spent on marketing, lobbying and management compensation.
I did not refer to the talent directly contributing to the technical progress.
P.S. - clarification: I mean not referring to talent at OpenAI.
And yes I have very little doubt talent at DeepSeek is a lot cheaper than the things I listed above for OpenAI.
I would be interested in a breakdown of the cost of OpenAI and seeing if even their technical talent costs more than the things I mentioned.
One example is that I've received offers to work in big tech in China at or exceeding my FAANG compensation here in the Bay Area. I have other reasons to believe as well but I can't talk about that in public.
> The numbers being thrown around are suspect, the paper they published didn't reveal the costs of all models nor the R&D cost it took to develop them.
did any lab release such figure? will be interesting to see.
In any AI R&D operation the bulk of the compute goes on doing experiments, not on the final training run for whatever models they choose to make available.