> But saying that we don't know how AI works is empirically false;

Your statement completely contradicts hintons statement. You didn’t even address his point. Basically you’re saying Hinton is wrong and you know better than him. If so, counter his argument don’t restate your argument in the form of an analogy.

> You'd think this, but it's actually wrong.

No you’re just trying to twist what I’m saying into something that’s wrong. First I never said it’s not deterministic. All computers are deterministic, even RNGs. I’m saying we have no theory about it. A plane for example you can predict its motion via a theory. The theory allows us to understand and control an airplane and predict its motion. We have nothing for an LLM. No theory that helps us predict, no theory that helps us fully control and no theory that helps us understand it beyond the high level abstraction of a best fit curve in multidimensional space. All we have is an algorithm that allows an LLM to self assemble as a side effect from emergent effects.

Rest assured I understand the transformer as much as you do (which is to say humanity has limited understanding of it) you don’t need to assume I’m just going off hintons statements. He and I knows and understands LLMs as much as you even though we didnt invent it. Please address what I said and what he said with a counter argument and not an analogy that just reiterates an identical point.