I made a quick video yesterday trying to explain what is the big deal with Deepseek.
There are many, but distilling OpenAI models is definitely not one of them.
Open-source is great, too.
But the real deal is them proving that with a new technique (MLA) it is more efficient (93.3% better) to run LLMs WHILE preserving same performance.
Let's use this post as a forum, ask any questions.
What do you think? What are you excited about? Concerns?