My concern is specifically that, if a human can't spot it, an AI-powered anti-cheat could be inadvertently trained on suspicious but legitimate games, leading it to flag legitimate players, which might trick a human overseer, which would then reinforce the inaccurate flags, creating a feedback loop.

The risk of false positives are quite high as once you're at the point where you need AI anti-cheat, you need it because humans are no longer effective at spotting cheating, so human oversight would likely not be all that useful. 2/2

My main issue with this idea is that all it takes to ruin the whole system is a few false positives and suddenly you have a feedback loop of legitimate playstyles being flagged, which are then used to train the AI further, causing it to target more legitimate players, repeat cycle. 1/2

Too direct, not enough brown-nosing.

An actual LLM would respond with "Wow! Your curiosity about whether or not 'we should do it' is inspiring! It demonstrates high critical thinking skills and a desire to wait for more information before making a decision. [proceeds to not answer the question]"

Ahh, that explains why the display isn't a CRT with so much burn in that you can't differentiate the difference between the reds and blues!

I think this is the first time I've ever seen a taiko machine/drum that isn't horribly yellowed and covered in a mysterious sticky substance with black shirt lint embedded in it.

Something, something, torment nexus