@Matthieu Crétier They are 'leaking' data that is uploaded and the prompt "engineering" info is also easily made visible. Honestly, if you are building an app, custom LLM seems to still be the way to go. Other thoughts?

I was not aware of the leaks. My main drive behind the question is the idea that if a GPT is well done and achieves 75% of what the standalone app is capable of, the user will see the convenience of just using ChatGPT Plus for getting stuff done. We are building a product aiming to enhance professionals’ productivity by delegating to AI, so sometimes I ask myself whether we are just wasting our time.

@Matthieu Crétier I stand corrected - I was told by a trusted source that the data leak into prompts is only for the non-api users. As for the question on wasting your time I see workflows, organization of data and versioning along with network effects as the primary moats available. Good luck! We are also in the enterprise productivity sector.