Tweet ghim
My thoughts on DeepSeek's impact to the market and scientific community:
BLUF: Short term de-risked into uncertainty, long term hugely bullish for AI and will cause mindshare to swing fully back from meme to AI.
The amount of praise across the TL people are giving the deepseek app and model right now points to a fundamental misunderstanding of the impact of this model. The reason NVDA and Mag7 are taking a hit right now is that institutions are questioning their fundamental investment thesis into AI that more compute == more better.
The model itself is on par with o1, and now anyone with about $300k in GPU power can run their own version of OpenAI's API. Crazy. Huge. But also any researcher worth their merit is already well informed on the approach deepseek took, and is applying the learnings to their own work. Soviets got Sputnik into space first but the US was the first on the moon.
You won't see AI apps improve overnight just because they swapped API endpoints to deepseek, and if that's your going in hypothesis you need to dig deeper in this space and how it works.
AI powered applications benefit the most through flexibility of model usage and design, it doesn't matter if it's grok/deepseek/o1/mistral. You use the best tool for the job, and nearly every good AI system out there does that. Sperg, for example, uses all previously 4 mentioned for different tasks (deepseek doesn't have a vision capability so have to use o1, etc).
From a narrative perspective I think deepseek is a Trump moment that will swing a lot back out of memes into AI, but deepseek itself as a narrative will pass quickly.
English
