What is the hype about DeepSeek

1. It is a very sleek application

2. Its results are very useful and perceived to be better by wide community

3. It is a open weight / open source model

4. Its performance on benchmarks is comparable to that of OpenAI o1 model

5. It costs mere $2.19 per million output tokens (30 times its cheapest rival o1 from OpenAI)

6. It required only $5.5M compute to train this model. This was the main reason of the selloff of stocks like Nvidia where everyone is speculating drop in compute demand

7. It uses Reinforcement Learning, upcoming technique, which picked up even more after the release and popularity of DeepSeek

8. It is from China closing in on the gap from US