DeepSeek V3 is BEST Opensource LLM EVER! Beats GPT-4o (Fully Tested) | Free AI Model

[@armauploads1034]

Thank your very much for this Tipp, like always your video is gold. PS: I suspect that inaccurate answers from AIs are often produced because their capabilities are deliberately throttled FOR THE FIRST QUESTIONS in order to - don't laugh - save computing power and also keep electricity costs lower. They often only make a real effort when explicitly asked for (Further questions with the explicit request for a better answer). I seem to have this experience quite often. I suspect there is a resource-saving system behind it.

[@yngeneer]

I'm afraid that this model is just trained on that -next-response-word-count issue, thats all. Did you heard about 'people don't think in tokens' ?

[@Pushkarkv]

Bro try o1 also