This was great. Finally, a sketchy use of LLMs I can get behind.
Computer science sketch comedy, so to speak.
The future certainly looks brighter with minds like tom7 bringing their talents to bear on alignment problems in machine learning.
Umm, this is outstanding. Tom7 reminds me a little of what I imagine would happen if Fabrice Belard took a bunch of mushrooms, encountered ego-death, and came back realizing not much mattered.
Now I'm going to go spend a long time wondering what other things could be hooked up to the badness metric in TeX algorithm land.