Can Assistants Work As Well As Custom GPT's?

API
1

I have created a custom GPT and an Assistant. Both are trained on the same 11 PDF files and both have the same instructions. When I ask the same questions to both of them they provide very different qualities of answers. The custom GPT is far superior in terms of its quality of answers.

I see some old posts late last year / early this year from people in this forum complaining about Assistants not providing as good responses as custom GPT’s. Is this still the case? Does it seem like something that will be solved? Does anyone know why the outputs would be different? Could it just be that I need to experiment with different models available to the assistant or different temperatures or other settings? Or is there something fundamental about how GPT’s operate that would make this not a solvable problem?

Also, it doesn’t look like a custom GPT can accept inbound API requests. Is anyone aware of any functionality that might be released in the future that would allow custom GPT’s to be able to be controlled via an API? My use case is to be able to create a website chatbot that would be able to respond as good as the custom GPT (either by using the Assistants API or by someone interfacing with a custom GPT - which currently doesn’t seem possible).

Thanks,
John

1 Like
2

Although they are conceptually the same I think it’s fair to assume that the ChatGPT version is a little more fine-tuned, receives constant updates, and also benefits from cost-hungry settings.

No, this is what Assistants are for. It may be that in the future they combine the two together so a Custom GPT can also be an Assistant, but there hasn’t been any word on this.

1 Like
4

Thanks. Are you aware of another AI platform that might be better suited to my use case than OpenAI? One where they expose their better version of the model through an API?