Domain expertise or just documentation?

I was talking recently to a person who has more expertise than me on developing AI agents for an industry specific domain task. While we were discussing on fine tuning LLM models, he said all we need is task documentation and the LLM will learn and do the rest without needing many inputs from human experts who are currently doing this work.

Fellow members, do you agree with this comment? Do we not need human domain knowledge to fine tune LLMs?

Can we get the same result by just uploading 100s/1000s of documents and start getting results?

2 comments