Yes, it offers a lot more than I initially thought. I think for the presentation they wanted to emphasize how simple it was to build and use, so I missed the implications of what they were showing. But the ability to upload a document base to one of these "GPTs" that it can then rely on for its responses means they must have some RAG system implemented on the backend, probably a pretty good one as I expect them to be very good at embeddings and search (they have a huge number of Google alums). It can also have integrated Code Interpreter, Data Analysis, and Image handling/generation if you tick the boxes, plus what looks like a very significant list of functions it can call in order to act as an agent (e.g. changing your calender entries or sending messages at your request after double checking, something that's probably going to give simonw a heart attack).
It looks like at base, with a short text description of what you want it to do and uploading of relevant documents, it may be able to replace like 95% of the GPT wrappers that have sprung up in the last 7-10 months. That's pretty impressive, because while all of those were doomed because they had no ability to defend themselves from competition, some of them were pretty useful.
It looks like at base, with a short text description of what you want it to do and uploading of relevant documents, it may be able to replace like 95% of the GPT wrappers that have sprung up in the last 7-10 months. That's pretty impressive, because while all of those were doomed because they had no ability to defend themselves from competition, some of them were pretty useful.