say you had very vertical trained models, such that you had like 1000 separate LLMs trained on specialized data and then others LLMs trained on which LLM is most likely to have the data you need, sort of like the way Wikipedia is interlinked, or hierarchical, or essentially like a db index, over nested LLMs, performance would scale higher with many more highly focused models, at least that's my understanding of what possible here.
Comments
say you had very vertical trained models, such that you had like 1000 separate LLMs trained on specialized data and then others LLMs trained on which LLM is most likely to have the data you need, sort of like the way Wikipedia is interlinked, or hierarchical, or essentially like a db index, over nested LLMs, performance would scale higher with many more highly focused models, at least that's my understanding of what possible here.