What happens when LLMs are prompted without fine-tuning?

Question

LLMs can be specialized to a task without any fine-tuning where the prompt acts as a form of in-context learning approach, how does that work and what seems to happen internally in the network?

In: 0

Answer 1

My understanding of the technology, and one of the “scary” things, is that those who created them cannot tell you how they do what they do. I don’t think there is currently an answer to this question.