Differences between knowledge data vs fine-tuning data

Could somebody emphasize differences between substance and formattings of

  1. Knowledge data for custom GPT
  2. Fine-tuning data

I’m interesting specially in:

  • Differences in formatting
  • Substantial differences

The background of my question: at the end of the day, both adjusting methods serve for the same goal: to give answers more exact. So there are more related questions rising:

  • Is it a correct approach: fine-tuning data are rather rules for answer data extraction and styling, knowledge data is a factual information, where answers are extracted (formulated on the basis of)?
  • Is it proper way to use same data for fine-tuning and knowledge?
1 Like

The Knowledge Base for a CustomGPT is the same as File Upload to an Assistant via the API.

There is no way to Fine Tune a CustomGPT via the ChatGPT interface.

Fine Tuning for Assistants is meant for cases when the System Message (the Assistant equivalent of the CustomGPT Knoweldge Base) would be too long otherwise.

Files uploaded to their respective knowledge bases can be in any format. Files used for Fine Tuning must be a .json in a specific format, that includes at least 100 Input / Responses to be effective, and should be used when you expect a specific response to a given input vs a creative response with variability.

1 Like