Similar to other reports, plugins in localhost seized to work, other prod plugins work fine.
Only localhost Dev Plugin
I think some messup happened on OpenAI front? my plugin model changed from ChatGPT to GPT4 and then local plugins just stopped working.
Logged out and back in, cleared cookies and cache and the issue remain the same, whever I ask my local plugin, ChatGPT just pretend that it doesn’t have any information after 2021.
Are there any status page or any updates from OpenAI?
And one more request please, if there is any maintainance, upgrades, outages that will impact users, appreciate if this can be posted somehow or send notifications via email to user.
Basically I was doing some overhaul to my plugin and really spent hours to debug it thinking it’s an issue with my plugin.
Only when I reverted to the last well known working plugin and it still didn’t work I realized that there should be something with the platform itself.
I appreciate the work that OpenAI do much but communication side can be greatly improved.
I believe I also have the same issue - the interpretation of the API seems to be working fine and generating the right request body, but then I never actually see a request made to the localhost server.
Last night it seemed to be alternating between refusing to use the plugin and making requests to the wrong endpoints, adding the http operation to the name (“/get_endpoint” instead of “/endpoint”)
As of this AM it shows it making the correct requests, but the requests dont actually get made (they dont show up in my server logs) and then hallucinating the responses
Is anyone with the ChatGPT Plus plan having that issue as well? I’m using the free plan and have the same issue. ChatGPT claims to send a request to my local server but actually doesn’t. It shows a loading spinner all the time.
Yes, I’m a plus sub, and it seemed to stop working after they enabled GPT-4 for plugins yesterday.
even existing sessions with old the model don’t know how to use the plugin any more.
Mine seems to be working fine in terms of requesting, but it is not adding the API response to the prompt so it is just inventing URLs and adding them to the list. Might just be a communication problem and the AI is just rolling with the punches.
But another big caveat, only GPT-4 can work with plugin, and with the the very limited cap currently it basically render plugins useless.
Understand the high demand for the awesome GPT-4, but basically the limits either need to significantly increase, or bring back 3.5 with plugins to continue the conversation when the the limit reached. right now if the limit is hit and model is switched, it doesn’t understand plugins any more.
So many kinks need to be worked out, love the work you guys are doing, but please be careful not to break the basic stuff that used to work, I’m sure you know that testing requires lots of back and forth!