News for hackers. Stuff that matters.

A comment on Interesting Interview with DeepSeek's CEO. Parent comment

Re:Interesting Interview with DeepSeek's CEO by arisAlexis on Tuesday December 31, 2024 @06:52AM (#42558169)
Of course if you arrive last and copy all the existing architecture you can train it cheaper
[ Parent | on HN ]
Re:Interesting Interview with DeepSeek's CEO by viraptor on Tuesday December 31, 2024 @06:56AM (#42558194)
No, you can only train at the same cost then. (Actually higher, because you don't have the existing hardware/power agreements) The whole point of the last model was that they made significant changes beyond just copying.
[ Parent | on HN ]
Re:Interesting Interview with DeepSeek's CEO by arisAlexis on Thursday January 02, 2025 @08:16AM (#42574131)
No because you can train once and avoid all the errors that are costly and make you train and retrain.
[ Parent | on HN ]
Re:Interesting Interview with DeepSeek's CEO by msp26 on Tuesday December 31, 2024 @07:55AM (#42558411)
> copy

You mean build on existing public research? Everyone does that. At least deepseek, meta etc. also have the decency to publish research back into this ecosystem.

[ Parent | on HN ]