Fewer than 4.5% of Microsoft 365 customers pay for Copilot after three years, only 1% use it weekly, and Microsoft raised prices regardless.
Fewer than 4.5% of Microsoft 365 customers pay for Copilot after three years, only 1% use it weekly, and Microsoft raised prices regardless.
You’re doubly confused. You’re thinking of GitHub Copilot, when this is about M365 Copilot. Last I heard, they have 78 different products called Something Copilot, so confusion is understandable.
But then what you’re actually thinking of is the “GPT” series of large language models, developed by OpenAI.
GitHub Copilot is merely a GUI and a harness for large language models. It defaults to the GPT models, and is probably somewhat optimized for them, but it can use other models as well.
This harness can influence the quality quite a bit, as it decides which source code files to feed into the model for context. I’m not aware of people saying that GitHub Copilot is particularly good at that, but it’s available as an extension for IDEs, so it automatically knows which files you’re editing, which can be useful.
And yes, the GPT models have fallen quite a bit behind since the start of the year.
But in Copilot Chat in MS Office for example, isn’t the LLM itself Copilot? That’s what I’m talking about as being a weak LLM — I know originally it was some variant of ChatGPT.
GitHub Copilot now is less a model and more a router. Kind of a pity, they had a solid lead there. I at least used it to complete yaml and json configs and it made fewer typos than me.