#Custom GPT vs Projects

1 messages · Page 1 of 1 (latest)

lavish spoke
#

Hey all, I'm curious of the following:

  1. which feature you currently favour between Custom GPT vs Projects, when you want to create a customized, application-tailored assistant? Why and how do you use it?
  • one clear pro for the former is the ability of Actions/API integration while the former can neatly compile history
  1. in either of the above options, how do you efficiently create and manage your knowledge base? While you can connect external repos like Google Drive in a single chat, you can only upload static, individual files in the custom knowledge base. Am I missing something here? For ex, Claude Projects can connect to these repos where i can more efficiently create knowledge base schemas (while updating them is still manual)
little harbor
#

I made an API that makes the model read again every message it receives (It's a single file, but still), you add a custom action

jagged nebula
#

I've often been wondering about which option was best myself. Always got a preference for custom GPTs for the convo starters, the fact we can assign a default model to them, the short description that can always be seen on top upon each chat, and the ability to set avatar pictures to your GPTs (very silly I know, but still. A shame we can't get such features in projects.)

#

That said, Projects allow to group chats within one specific space, which is very handy - Open AI should have implemented a category system since like forever. Plus, you can access to your saved memories OR those saved within the project only. And I just found out you can edit projects instructions on the mobile app, which is impossible for custom GPTs. It's painfully obvious that Open AI decided to focus on projects, to the detriment of customs. So I think I definitely will use them more often in the future.

As for knowledge base, I just input whatever info I want in a notepad file, with the specific instructions I want chatGPT to follow for this doc, and then upload it in there. I wish there was a simpler and more dynamic way to input instructions & infos in the system, though; like inside a Canvas file that acts as memory, and that persists over all chats that we and ChatGPT can access and edit anytime. It would be so awesome!

halcyon elbow
#

i havent made a custom GPT yet. What I've been doing is organizing all of my old chats into projects based on topic. The difference for me when/if i decide to make the type of focused assistant you're talking about would be, which one can accept the most uploaded files and/or access my google drive files etc for as much context and reference storage as possible, and also would need cross chat memory within the assistant