DeepSeek AI

Published: Jan 29, 2025

Last updated: Jan 29, 2025

DeepSeek-R1

Chinese startup DeepSeek rocked the AI industry by releasing their new open-source LLM, DeepSeek-R1. Quite comically, the DeepSeek hype started right after the US announced their $500 billion Stargate AI project. DeepSeek has already caused massive panic with investors in the AI and tech industry catalyzing a selloff Monday morning (1/27), resulting in up to a 13% price drop in Nvidia, nearly $1 trillion loss in tech stocks, and 3% drop in the Nasdaq.

One look at Nvidia's chart and it's clear there are fears of disruption.

Screenshot of NVDA stock chart spanning January 23 though January 28, with a sharp drop on January 27th

It would be an understatement to say that US tech companies are shitting bricks right now.

Why is DeepSeek so hype?

Due to export restrictions from the US, China isn't able to buy the same high powered GPUs and hardware available to US based AI companies. As a result, DeepSeek optimized for extreme computing and cost efficiency. To analogize, DeepSeek is a souped up Honda Civic with a laptop, while ChatGPT is a fully loaded Lambourgini. DeepSeek-R1 is so efficient it's become the best open-source model and is comprable and in many cases better than proprietary models such as OpenAI's o1 and Meta's Llama3. In terms of cost, DeepSeek is objectively and by a HUGE margin much cheaper than any other AI competitor to train. Sam Altman stated the cost of training GPT-4 was more than $100 million, while DeepSeek R-1 costs less than $6 million, only 6% of the cost. Keep in mind, this is the cost to train the model and doesn't factor in research and development.

In addition to DeepSeek-R1's extreme efficiency, the startup released the LLM under an open-source MIT license and released a technical report detailing the training process. The technical report has some banger graphs that illustrate it's performance in contrast with competitors such as OpenAI's o1.

A series of charts showing how hard DeepSeek-R1 shits on OpenAI o1

With DeepSeek being open-source and highly efficient, it's possible AI researchers and developers will pivot away from proprietary offerings in favor of DeepSeek. This could prove to be a massive shift in technological dominance away from the US to China.

Additional thoughts

I have always kind of thought that AI was over-hyped and over-valued, especially when you consider how much money is being pumped into it by big tech firms. Don't get me wrong generative AI is a handy tool and has it's uses, $500 billion though? Are you kidding?? DeepSeek being able to completely leap frog the entire competition (seemingly overnight) with a sliver of the cost really looks sus for US tech. Where's all that money going? Why does it cost so much to produce an inferior product?

On another note, this release also shows how ill-informed investors are. Do they not realize DeepSeek also runs on Nvidia?

TL;DR

Chinese startup DeepSeek released an open source LLM that absolutely shafts leading US big tech.

References