Right, but these are general language models, just big ones - and they seem to be progressing quickly. People have started working on more specific systems: https://www.deepmind.com/blog/competitive-programming-with-a...

Once you have something, you can iterate on it, raising the bar. It seems to be accelerating and doesn't feel like it'll be that long before the bar get's raised over our heads.

Github Copilot writes 50% of the code I write now, perhaps more. I'd say that's actual work.

You still have to prompt this code and you still have to verify the code works and fix it when it doesn't. People who use more basic autocomplete probably have it "writing" a double digit percentage of code, I don't think that makes it fair to say autocomplete is doing a double digit percentage of the work.

What type of work do you do?

Lot of human work is just "variations on common templates".