I’ve created a facebook messenger bot that’s part of my friends group chat and we had a blast using it with text-davinci-003 model. I’ve rewritten the implementation to utilize gpt-3.5-turbo now which works well
The comment “The name of the user in a multi-user chat” implies that gpt-3.5-turbo might support group chats in the future (or it does already?)
So my question is, did anybody had success in running gpt-3.5-turbo on a group chat? Any advice regarding prompts or implementation? Our case is bunch of people talking away and occasionally wanting the bot to join in the conversation. We also have a “secret word” that resets the conversation. I’ll be thankful for any clues.
Thanks, replies under your post summarize that “name” param is internal for OpenAI staff to monitor abuse, and not for end users to use. We’ve tried to play dungeons and dragons gpt-3.5-turbo being the game master with a prompt that instructed that there are multiple people on the chat and all messages will be preceded with "Name: "
Some of the results were very promising, but I think there’s way more work to do regarding prompt engineering.
For instance, we’ve instructed the GPT there are 5 people on the chat (giving their names), and we asked to create the characters for the game. I’ve told I want to play some mage, and my other friend chose some other profession. GPT asked what other three players choose. I’ve told it they are not playing. It decided to start the game then, which was super cool.