Andy Halliday

And the wisdom of crowds is happening inside a reasoning model.

They're sort of creating separate argumentative entities inside the model in order to accomplish a conversation that actually improves their reasoning accuracy.

Wow.

Isn't that interesting?

Wow.

I love that that's what's happening inside this vast and generally inscrutable collection of simulated human neurons inside a neural network.

And just to toss out some terms, like there's some, you know, dialogue in the chat stream here about LLN Council, you know, the idea of having, you know, directed, you know, through prompting, direct, you know, a model to take multiple positions, right?

So in effect, create positions.

representations inside your output responses that represent different points of view.

That's all part of this thing that's related to this idea.

And mixture of experts is really what that is, right?

If you can generate conversation among a range of experts, each of which has a unique focus,

provides differential perspective, which came from that abstract, by the way, as a term.

All of these things are related and bring it forward to what we expect will happen, what I expect will happen, is that artificial general intelligence will reveal itself in AI by a council model rather than as a single unified intelligence that

that is, you know, monolithic in that respect.

It's going to have very, very broad and multiplexed capabilities of reasoning that provide for point counterpoint, you know, thesis and, you know, rejection or proof and so on.

All of that is all part of this thing moving forward.

And I think you can probably observe that in the,

structure that axiom math is using in order to get to really advanced mathematical theoretical reasoning.

And that strategy, by the way, it seems focused on just making more efficient inference, right?