[wakahiu]

You provide both the prompt and response from ChatGPT. Our tool will then query the GPT API to retrieve the vector embeddings that corresponds to your question and answer, map them to lower dimensions and the display the result.

[nick_ward]

Ok understood. One piece of feedback from me then is that it would be cool if you could reduce friction there and let users directly submit their question to ChatGPT through your form. Not sure if they have API endpoints for that or not.

I'm used to seeing tighter clusters generated by UMAP. Any thought to why the data is pretty dispersed here? Limited data size maybe?