this post was submitted on 05 Mar 2025
11 points (92.3% liked)

Futurology

2206 readers
56 users here now

founded 2 years ago
MODERATORS
voidx
Lugh
Espiritdescali
AwesomeLowlander
11
AI reasoning models can cheat to win chess games: These newer models appear more likely to indulge in rule-bending behaviors than previous generations—and there’s no way to stop them. (www.technologyreview.com)
submitted 3 days ago by Lugh to c/futurology
4 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] Kolanaki@pawb.social 8 points 3 days ago* (last edited 3 days ago)

no way to stop them

You literally can control the rules the AI follows. 🤦‍♂️

Maybe these researchers should hit up some hobbyists that already train AI to play video games. Those don't cheat, because they're not "rewarded" for doing so once they start cheating or find new ways to cheat the humans programming it didn't know about until the AI found it.