I know of no such thing, and google doesn't turn up anything in a cursory search. Modresolving NO.

@metacontrarian Github Copilot only has 1.3 million users, 10 million people seems steep in comparison.

Fine-tuning on personal text is still very cool and doable, but I think the user adoption requirement sinks this.

if you wanted to do this right now, how would you do that?

How does this resolve?

Utility-wise, this has been true for some time already, as there's ample examples of davinci/turbo utilizing private data as a knowledge base by means of embeddings, often using libraries like GPT-index and langchain that fit this exact purpose very well.

If the exact criteria is fine-tuning, then that has also happened since OpenAI has exposed a fine-tuning API for their non-chat models for a while, which i would argue in the vast majority of cases is being used for private data.

If the market demands a specific product connecting to APIs for Slack/Whatsapp/Discord/Emails etc., AND demands fine-tuning, then i think the probability is close to 0% considering the ineffectiveness of that approach compared to context-length increases combined with embeddings.

@minosu Re: Para 1 -> "fine-tuned on people's personal texts"

Re: Para 2 -> It has to be fine tuned on people's personal data and the application doing that is popular - which i dont think has happened.

@firstuserhere In that case, that's a no for me. :)

A bunch of similar markets (Differences: Timeline, what counts as popular)