Post #1490134
2026-02-26 13:24 UTC
@ngaylinn This might be a disambiguation issue. What you're describing sounds like the old symbolic RL, where you get proofs and guarantees.
What is called RL here is really just a gradient estimator, so you can backpropagate through a sampling process. There are no explicit states or modeling of the problem. Just an action space and a reward signal.
I can certainly understand your discomfort. It's completely black box, and people are very rapidly putting it in charge of things.
Replies (1)
-
@ngaylinn@tech.lgbt 2026-02-26 13:29
@pbloem I understand these different kinds of RL, and my point is how far we've ventured from a place of confidence and safety. Yes, the key issue is handing decision making to a black-box oracle. Especially one whose primary "experience" is text from the web. That's really not the bias I want in such a system.