What an awesome talk given by Luke Demi on the do’s an don’ts of fine-tuning LLM’s at this month’s Lancaster AI meetup.
The important take-aways included fine-tuning is really only for changing the interface of the chat, so things like personalities or coding. It did poorly if you attempted to train it on new data. Luke showed using his personal journals how this would work.
However, using a RAG in combination with the fine-tuning was more effective, and it had is voice, and could pull and generate similar data.
Thanks again to all who came out today!
