this post was submitted on 15 Jun 2026
25 points (100.0% liked)

technology

24389 readers
206 users here now

On the road to fully automated luxury gay space communism.

Spreading Linux propaganda since 2020

Rules:

founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
[–] yogthos@lemmygrad.ml 4 points 3 hours ago

It's not an LLM, because the model is not trained on text tokens here. The weights of the model encode a physics simulation that models the world to some level of fidelity. And that's what gives the model actual reasoning powers in a human sense. The LLM layer could live on top of this model, but you might not even need it because you could potentially take this base physics simulation, and then train the model to understand language the way you teach a child. At that point its understanding of language is grounded in physical reality which is what gives us all shared context. The whole reason we are able to communicate effortlessly is because we all have a roughly similar underlying model of reality, and we can map language to it. But the problem with LLMs is that they're just stochastically string tokens together without any grounding.