Idea from 08/09 meeting was to perhaps log some slack reactions. Potentially, we use this dataset to fine-tune (using RLHF or similar) a model in the future.
Some things to consider:
We easily can get the query (or chat history) and the response, but maybe for this, we want to log the full prompt (which includes the context provided). Is the full prompt required? Or is the query / chat history and response good enough to fine-tune?
Idea from 08/09 meeting was to perhaps log some slack reactions. Potentially, we use this dataset to fine-tune (using RLHF or similar) a model in the future.
Some things to consider: