deepseek-ai/DeepSeek-V2 · This is by far the best model I have seen until now.

After testing many different models (gemini,gemma,llama and claude) I think this is the most promising model so far regarding reasoning and logic.
I wish to contact the people that created this model, because I have some ideas that could make AI progress by a factor of 2 or 4. Meaning: a smaller model could have the same reasoning capabilities.
I am serious, the problem is not just training (which can be done in a different way),
but is quite complex and I wish to brainstorm with the team and see if my ideas are right. If they are, and I have no reason to think otherwise, we could make a big breakthru in this field.