search

yGuy / chatgpt-mattermost-bot

A very simple implementation of a service for a mattermost bot that uses ChatGPT in the backend.
MIT License
145 stars 50 forks source link

Once ChatGPT has been tagged in a thread, it will automatically reply to any reply.... this is very annoying. It should only reply when it is tagged in the latest reply of a thread. #42

Closed OlivierVernin closed 11 months ago

OlivierVernin commented 1 year ago

See the attached screenshot, as soon as i mentioned @chatgpt once as a string (not even a tag) on the red cross. he started to pollute my conversation with my team. This is especially annoying as my team would like to discuss the response from ChatGPT in the same thread. For us, ChatGPT should only been triggered on the reply where he is explicitly mentioned.

Would there be a way to have different mode of triggering chatGPT sometimes in full interactive mode (as now) and sometime in a more on-demand way. This is especially true for ChatGPT4 with an API limit of 24 requests per 3hours, you don't want to trigger him at every reply. Still when summoned on-demand ChatGPT should be able to use all the previous messages of the thread.

Capture d’écran 2023-06-13 à 19 42 12

yGuy commented 1 year ago

ChatGPT is lying, here, obviously.

There is somewhat of a (very quiet) discussion over here: https://github.com/yGuy/chatgpt-mattermost-bot/pull/15 I am not sure about the usability, for those cases where you do want do have a dialog with chatgpt, I think it is pretty annoying to always tag it in every sentence. You don't do it in you example, either. I would love to see this discussed and see different alternatives. I know, the current situation can be annoying. You need to start a new thread in order to "stop" chatgpt from listening. We might come up with a stop-word approach. I don't like it having to @-tag it in each and every reply.

How would you like to see it implemented? Have you tried the version in the pull-request?

OlivierVernin commented 1 year ago

Hi, thank you for your quick response.

No, we are using the main branch version. What is different with the PR version?

Being able to stop ChatGpt on a thread seems a good compromise. As a stop word: "@chatgpt stop". Then a simple "@chatgpt" would resume.

Side question, when tagging ChatGpt, he only reads the last reply and not all the thread before? It could actually be nice to ask ChatGpt his opinion regarding a actively debated discussion that happened before in the thread. Like "@chatgpt please summarize the above thread"

yGuy commented 1 year ago

Regarding the differences, just look at the PR and the discussion.

Regarding your second idea and question: See also https://github.com/yGuy/chatgpt-mattermost-bot/issues/28

yGuy commented 1 year ago

With the 2.0 release, you can silence the bot and later make it respond, again. Can you test the version and provide feedback? That would be awesome. Just tell it to shut up or leave at any time and it should wave good-bye until you reactivate it.