Many users call the proof that latent world models need exponentially less data than LLMs a very interesting result with strong theoretical justification for prediction in representation space, while a few dismiss predictive AI as insular.
5 comments with sentiment.