3 ms·
Delivering Copilot for Everyone
- giantg2 3y agoThey should have called it Clippy.
- istillwritecode 3y agoAs far as I can tell, all copilot does is to regurgitate web results it has found, often out of context.
- poisonborz 3y agoMy experience as well. They try to do this as cheaply as possible of course, but this can be felt strongly compared to "real" tools.
- coolgoose 3y agoI did have one cute moment regarding the copilot integration. I downloaded a menu of a restaurant in Dutch, and it could translate it and explain what those dishes are and what was most likely to be similar to chicken nuggets.
- infecto 3y agoI have not been happy with the Microsoft integration so far. I think part of the problem is it’s quite hard to use current LLM in very generalized ways. Freaking amazing when used in a specific workflow but when you are just doing some generic task in office or elsewhere, it can be exceedingly hard to build the right context for it.
- nbittich 3y agoNo thanks. Doing my best to get rid of everything faang or Microsoft sold me in the last decade, it feels good thinking nobody is spying on me in order to make sure they can sell me a casket when I get diagnosed with cancer or smth
- CrypticShift 3y agoI'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.
- judah 3y agoLaunching 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?"
- edwhitesell 3y agoThe description of the App in the Google Play Store lists GPT-4 & DALL-E 3.
- vitorgrs 3y ago10 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.
- pupileater 3y agoIn the video the prompt asks the Copilot to "generate a logo for my fantasy football team" then it goes and generates a typical "cute cat with football helmet" image, heh.