What standardized dataset formats are people using?

[norton120]

Describe the feature or potential improvement

I'm thinking about how the dataset editor could be more effective, especially for members outside the immediate development team. As pure json (even with a json editor or like Ace or similar) it is still pretty unwieldy.

With a common standard we could present an (optional) first-class GUI - for example, if using the OAI schema the input and expected output messages could be formatted to look like a chat (with different colored bubbles for user/assistant) and be editable in-place. That would be a much lower barrier to entry for a CTO who has a specific example she wants tested (which may not be ideal academically, but happens quite a lot in practice).

maybe even do the schema detection on the fly - so if the input/output conforms to an editor we've built, we just default to that editor, otherwise show the raw json editor. Users can always toggle.

Additional information

No response