I don't like #AI. I think it's problematic in a lot of ways, that I won't repeat here as I believe you heard these (very much true) arguments 100 times.

But #DuckAI helped me quite a lot yesterday, and I don't know how to feel about that. By curiosity, I made it write #elisp code to create new useful commands for #emacs which will help me in my everyday work (for example `C-c m` now creates a PDF using `pdfmom -k` so I don't have to use a terminal at the same time when I use #groff, `C-c t` adds `/|` before some punctuations signs following French typographic rules, …). These functions are very simple, most of you here could have written them by hand in less time than it took me to write the prompts, but I'm no developer, and don't have the time to become one, unfortunately. They work as intended (I can control that as, again, they're simple things), and I more or less understand them, as the #LLM described them. I even asked it questions to understand some parts of the code, and it answered (correctly it seems; I checked and as far as I can know, its answers seem legit), so I was able to modify them by hand to fine-tune them. One was wrong, but when I sent it the error message, it corrected it and explained the error to me.

It didn't feel like I was relinquishing brain power; in the contrary I learnt things. I could have done that by hand, and I would have learnt far more, but it would have taken months and I don't have months: a bird in the hand is worth two in the bush.

I still think that LLMs are a problem for our societies, but even more so than I thought as it seems to be actually useful in some cases.