IMO people are lost on the text component and missing the means by which LLMS co

IMO people are lost on the text component and missing the means by which LLMS construct ‘meaning’ out of relations just as our neurons do. The question since the beginning is whether these things can create a world model, that like Tesla’s is operational. But tesla’s world model is concrete while the conceptual world consists of abstractions. But it turns out that these two geometries (and they are constructed as relational geometries) are compatible.

The reason my generation couldn’t pull it off, was the availability of compute. But the fact that we could brute force world models through language (which in retrospect should have been obvious) was just something we didn’t predict.

That’ the leap. That one leap save us decades.

Reply addressees: @CloudByter