You say that, but it's been better than most employees for a year or so ( *for specific tasks, of course. It's still not better than "an employee" )

Not true at all with frontier models in last ~6 months or so. The frontier models today produce code better than 90% of junior to mid-level human developers.

Just like a real employee!

And just like a real employee, this makes it work worse.

(Old study, I wonder if it holds up on newer models? https://arxiv.org/pdf/2402.14531)

Interesting, I've actually found swearing at the dumbass bots to give better results, might just be the catharsis of telling it it's a dumbass though.

There's certainly catharsis, but I'm taking the opportunity to train myself to be less frustrated by apparently stupid responses in the hope it makes me a more effective communicator towards humans.

My worst experience of co-workers (and in contract work, customers) is when they treat me the way I used to treat ChatGPT. If I can make sure I myself am never like that, just by training myself to different behaviour through my use of the infinitely patient LLMs, all the better. :)