I think it's already out of date with verifiable reward based RL, e.g. on maths domain. When "correctness" arguments fall, the argument will probably just shift to whether it's just "intelligent brute force".
I think it's already out of date with verifiable reward based RL, e.g. on maths domain. When "correctness" arguments fall, the argument will probably just shift to whether it's just "intelligent brute force".
The set of tasks for which "correctness" is formally verifiable (in a way that doesn't put Goodharts Law in hyperdrive) is vanishingly small.
"stochastic genius"