I tried this in branch gpt-3.5-turbo and I was hopeful. I was looking forward to sending through the last few messages in the conversation in a sensible way.
Ultimately, although I think it has potential for anticipation and salience, that model is really no fun to talk to. Using 'System' and 'Assistant' configuration binds the conversation to a very, down-to-business,
"I'm an AI chatbot here to serve you. No, I don't want to chat about 'why the sea is boiling hot or whether pigs have wings'. What are your action items for the day?"
kind of attitude. Even after testing the configuration in different ways - jacking up the temperature and/or top_p - talking to it was like licking a brick.
I'll probably revisit this at some point, but for now, it really sucked the fun out of the project.
I've included the other code changes that are now in the current dev branch and am continuing to work out kinks and will proceed with other gh issues for now.
I tried this in branch
gpt-3.5-turboand I was hopeful. I was looking forward to sending through the last few messages in the conversation in a sensible way.Ultimately, although I think it has potential for anticipation and salience, that model is really no fun to talk to. Using 'System' and 'Assistant' configuration binds the conversation to a very, down-to-business,
"I'm an AI chatbot here to serve you. No, I don't want to chat about 'why the sea is boiling hot or whether pigs have wings'. What are your action items for the day?"kind of attitude. Even after testing the configuration in different ways - jacking up the temperature and/or top_p - talking to it was like licking a brick.
I'll probably revisit this at some point, but for now, it really sucked the fun out of the project.
I've included the other code changes that are now in the current
devbranch and am continuing to work out kinks and will proceed with other gh issues for now.