On Feb 26, 2024, at 4:31 PM, Karl Benedict <[log in to unmask]> wrote:

> Eric - it sounds like we may be at about the same point: I am wanting to start working in the area of fine-tuning, specifically focusing on Chat-GPT generated data management plans that would then be revised by experts and used as a fine-tuning data corpus for (hopefully) improving the draft DMP language provided by Chat-GPT. This is part of a broader experimentation with DMP generation prompts derived from machine-readable DMP content.
> 
> -
> Karl Benedict
> Director of Research Data Services/ Director of IT
> College of University Libraries and Learning Sciences
> University of New Mexico


Karl, yep, the process of fine-tuning seems to be one way make better use of large-language models (LLMs). Yesterday, I received a generic email message from OpenAI, and the message described how to fine-tune their models. I read it with great interest.

Others think the implementation of retrieval-augmented generation (RAG) is the way to exploit LLMs. It is lesser expensive in the long run, provides immediate feedback, and ensures your content takes precedence in the result. On the other hand, RAG requires a lot of prompt engineering and the pre-creation of some sort of index.

Thank you for sharing.

--
Eric Morgan <[log in to unmask]>