That exains a lot...

I do use AI to assist my programming, but I always take what it suggests as likely highly flawed. It frequently sends me in the right direction but almost never is fully correct. I read the answers carefully, throw away answers frequently, and never use a solution without modifying it in some way.

Also, it is terrible at handling more complex tasks. I just use it to help me construct small building blocks while I design and build the larger code.

If 30% of my code was written by AI it would be utter trash.

we can tell

That’s like, not a good thing dude

I mean, autocorrect has written about 50 percent of this comment, but that doesn't mean my phone is writing it for me as much as it accelerates what I wanted to type in the first place.

Maybe that's how they got to a 30% estimate, cause ain't no other way that would have worked

I wonder how they measure that. Writing 30% of loc with AI seems like it will be terrible. Writing 30% of each loc with AI (i.e. autocomplete) seems feasible

I could believe this.

But how is their OS or any other product even close to functional at that point?

Given they made a big hoohah a few years ago of getting rid of most of their QA—this in combination is a particularly bad look

I wonder if Microsoft has a model that learns from their own source code. I have tried having copilot write sql queries and they work (without modification) about 10% of the time

Pathetic