⚡ 4 h early
Adversarial Policies Beat Superhuman Go AIs
5 experts are actively discussing the implications.
2 experts
1 community
1 sources clustered
“These are very odd failures! arxiv.org/abs/2211.00241”
5 experts discussed this · 14 posts
Eugene Vinitsky: Taking more seriously the claim that recent ML models do not "reason", it still is quite odd the particular ways that superhuman game-playing models fail, suggesting that what they're doing is some…
Eugene Vinitsky: These are very odd failures! arxiv.org/abs/2211.00241
tweety fish: this is absolutely my hobbyhorse but nothing I've seen yet has convinced me that we do a good job of evaluating how the sort of solution manifold of LLMs diverges from that of humans and that strik…
Open the full discussion →