I recently realised that the Assistant API does not have the same capabilities as the Custom GPT, which can be accessed and created through the ChatGPT interface.
I have loaded the same training documents into a Custom GPT instance and also via Assistant API into a GPT-4 instance, but I have noticed a huge difference in quality. I got much, much lower quality answers to the same questions via the Assistant API than from Custom GPT.
Do you know what could possibly be the reason for this?
Do you know how to give an Assistant exactly the same capabilities that a Custom GPT currently has?
If not, do you know when an API for the Custom GPT model will be available?
I am finding the same issue. Especially for pdf files that I upload on Assistants it refuses to process those documents. On the other hand, in Custom GPT it is able to do everything seamlessly. There seems to be a lot of smarts built behind CustomGPT which make the experience much better. However the issue is that CustomGPTs cannot be ported onto your own frontend.
I think Assistant API is interesting. I’m still trying to find the best use case for this API (I haven’t found the best suitable method for it yet). If anyone finds a solid example, I would love to check it out!
Interesting. I’ve noticed the opposite but my circumstances may be different. I put all of my training documents into raw txt files as they are much much lighter/smaller than PDFs files. The responses are always 99% accurate in the Assistant compared to like a really low 35% on the equivalent GPT. Only difference can think of is that my training data is HUGE. like thousands and thousands of pages. I think since GPT is web-based it makes it harder to do that many files but I don’t know.