Melanie Mitchell backs claim that today's AI 'are not LLMs' anymore

asusarla · x · 2026-09-27

Computer scientist Melanie Mitchell amplified Aidan Clark's provocative claim that "what we have now are not LLMs": she reminds everyone that a language model is a model of language, and today's AI systems — though built on pretrained language models — have drifted far from being purely language models after all the post-training. The exchange adds fuel to the ongoing debate over whether frontier models have outgrown the LLM label.

Related event: Melanie Mitchell Sparks Debate: Are Today's Systems Still LLMs?(21 posts)→

Original post →

More from Models

Models channel →