Categories
Me Myself&I

DeepSeek

Exciting. Lame as hell. Just as I thought it would be.

Exciting:

Doing decent LLMs output with a lot less is quite a feature. Mad props.

Lame as hell:
All the dudes online thrashing each other over how NVIDIA’s stock is a bubble-I-told-you-so, or how “the Chinese” are “kicking ass” or how US AI companies are crying etc.

Just as I thought it would be:
Maybe not that accurately! But my guts were telling me that there was a huge amount of optimization that could be done VS scaling up hardware and building power plants day in day out to improve AI’s performance. High-Flyer just proved it.

So to me the conclusion of DeepSeek is this: some of the brightest minds on earth, everywhere, are working really hard on AI. Progress is being made and shared at a fascinating pace. Hardware is hitting physical limits (there are rumors of the latest NVIDIA cards crashing a lot because they run superhot) so the possibility to now do more with what we have, is very welcome.

One interesting angle: the Chinese team had to find ways to optimize their systems, as they didn’t have access to the latest hardware. AKA, you get creative when you face constraints. AKA, blocking GPU export to China made them stronger.

lmao

Leave a Reply

Your email address will not be published.