What are custom GPTs build upon?

[a.van.hengel]

Hey there,

I was wondering if it is disclosed anywhere on what version of ChatGPT (or any GPT model) custom GPTs are build. And if you have a GPT, does it keep updating when new versions of GPT4 are laid out?

Thanks in advance!

[thinktank]

I’m not sure anyone has any clue what model is actually being used for the ChatGPT interface at any given time. Probably GPT4, just like they say, but who knows?

I assume CustomGPTs run on the best model available, since there isn’t an option to decide which model you access. When the model that runs CustomGPTs are updated, so will the backend of your CustomGPT. But, anything you’ve made—Instructions, Knowledge Base, Actions—won’t change and you have to change them if the update necessitates it.

However, that’s not something they do often or lightly. The latest big shift was from Plugins to CustomGPTs. The next might be the upgrade to GPT5, or maybe an innovation with Actions.

[_j]

@thinktank

However, that’s not something they do often or lightly.

ChatGPT is also the playground for testing new unrevealed AI models on trial groups and seeing the “which of these is better” results. Another part of having no guarantee of what versioned model you are interacting with.

GPT-4-turbo models on the API are basically a side product of making a ChatGPT AI that chats “efficiently” as its purpose, one can observe by the changes in quality in API models.

[thinktank]

That’s definitely true, and one thing I like about the ChatGPT UI is that you can wax rhapsodic with the prompt and not have an anxiety attack about the efficiency of the prompt and the number of tokens you’re using.

(Tried a one-shot within an existing conversation.)