I'm confused: Is this GPT-4 or GPT-3.5? The last reasonable answer I found (I forgot where) mentioned that they only downgrade it to v3.5 during peak hours. Still, even if it is v4, it seems to behave differently. The copilot should somehow be added independently to the LLM leaderboards.
back
3 comments
10 days later answer, but Balanced runs "multiple models", there's a decision making system which choose which model to use. It's mainly focused on search, and being faster to answer. Deucalion is a new MSFT-trained model that's being used here. Cheaper to run, too.
Creative and Precise runs GPT4 full time (not turbo), but both are still finetuned versions of it. To be more creative, and the other to be more "precise". They are slower.
Launching Copilot on my iPhone, I see an on/off switch "Use GPT-4", with it defaulted to off. Switching it on, Copilot says, "Nice! It will be fun with GPT-4 . Responses may be slower while being creative. What do you want to explore today?"
The description of the App in the Google Play Store lists GPT-4 & DALL-E 3.