Hinton: Large Models Do Actually Understand

Chris_Armstrong · x · 2026-07-19

Hinton has publicly stated that large language models "actually understand" language, rather than functioning as simple autocomplete tools.

The recap emphasizes that to predict the next word accurately, a system must understand the sentence in some sense. This raises a broader question: humans may eventually only be the second most intelligent beings on Earth. The post frames this as a direct challenge to the debate over "whether machines are truly thinking or if we simply misunderstand what thinking is."

Related event: Hinton Discusses LLM Understanding and Superintelligence Risks(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →