The discussion revolves around the availability of GPT-4 with vision capabilities in the Discourse AI plugin. tpetrov questions whether the gpt-4-turbo model includes vision, referencing the OpenAI API documentation, which suggests that gpt-4-turbo with vision is an enhanced version of gpt-4-turbo-preview with vision capabilities. sam clarifies that currently, gpt-4-turbo does not include vision, but the team is exploring integration via a pull request (https://github.com/discourse/discourse-ai/pull/470) to enable image captioning using vision APIs. The conversation also touches on the potential of an
In the AI plugin settings I can see this list of AI bots I can enable:
I don’t know the answer, and yet Vision is same as 4, it has a lot of issues and can be problematic. But AFAIK there should be own model for Vision and Discourse doesn’t use its own naming style.
FYI this is supported in Discourse Chatbot 🤖 (for “Basic” mode)
(Without RAG that makes it a bit toothless; otherwise it gives a lot of fun)
Please call Sam Altman 
, he needs to add functions to the vision model.
At the moment gpt-4-turbo == gpt-4-0125-preview aka no vision.
@keegan is exploring some vision integration here:
Particularly adding captions to images via vision apis (both open and closed models)
I do really like the “demo” power of … starting a PM with bot, uploading an image and asking about it, but need to think through flow cause we would need to switch in and out of models.
Maybe a concept of “auto tool” can work here, so if we see an upload we automatically route to tool and then back to the main model… will have a think.
Technically wiring this up is not too complex within our architecture.
Thanks Sam!
According to Vision - OpenAI API
Vision and turbo should be mostly the same (not sure though, haven’t tested thoroughly yet).
