I think we need to start thinking about one shot training. I.e instead of context into LLM, you should be able to tell it a fact, and it will encode that fact i

I think we need to start thinking about one shot training. I.e instead of context into LLM, you should be able to tell it a fact, and it will encode that fact into the updated weights.