Why did we decide LLMs are the de facto models for everything? AI has always been best when engineered for highly specific tasks. Is it just a data / training problem?

The path to better AI isn’t GPT wrappers, it’s making it as easy as possible to modify the underlying architecture for a specific domain and train it on whatever data you’d like.

Then, build agent networks to connect them.

LLMs are not the only solution for AI- focus on domain-specific models and customizable architecture for better performance