This may or may not be a thing, could be trolling too. However having Copilot on the computers at work and testing it out on mundane things like email summaries and replies, it’s pretty terrible. So if a LLM fails at simple language tasks, numbers are going to be spectacular. I was honestly surprised how bad Copilot is, having used some others, including smaller local LLMs. Maybe the enterprise version is a 1-bit.
This may or may not be a thing, could be trolling too. However having Copilot on the computers at work and testing it out on mundane things like email summaries and replies, it’s pretty terrible. So if a LLM fails at simple language tasks, numbers are going to be spectacular. I was honestly surprised how bad Copilot is, having used some others, including smaller local LLMs. Maybe the enterprise version is a 1-bit.
It’s not too bad at predictive text … seems like a lot of effort for something my phone can do locally though …