News for hackers. Stuff that matters.
Re:AI Models Can Provably Hide Arbitrary Capabilities by fenazego on Tuesday October 13, 2026 @02:53AM (#50067764)
Without having read the article, I would say it demonstrates that benchmarks alone, without knowledge of how a model was trained (c.q. constructed), cannot reliably determine whether or not a model has certain capabilities.
Re:AI Models Can Provably Hide Arbitrary Capabilities by ag8 on Tuesday October 13, 2026 @03:05AM (#50067855)
Yeah. Also, we think that in certain scenarios, it's possible for "cryptographic patterns" to arise in model weights via normal training, especially if models are trained against things such as honesty probes.