I'm also working on a similar feature, it's simple but complicated too "merge" information in when using GPT's. I imagine you've encountered it too. If I say I have an array of interests ['music', 'gardening', 'racing'], and a GPT replies ['music', 'gardening', 'racing', 'competitive driving'], you don't want additive synonyms. And then more complicated is deletions, if it replies, ['gardening', 'racing', 'skydiving'], does that mean I want to remove 'music' or did it just forget to include it in the response so should I just merge the old and new array.
I'm probably going to open source a llm-structured-data-merge-thingy library (that has different merge strategies) in the near future.
I'm probably going to open source a llm-structured-data-merge-thingy library (that has different merge strategies) in the near future.