What Yann LeCun said, newest first. Changes are found by AI and checked before they show; every quote opens at its moment.
2026-10-01 · World Governments Summit · 05:48
“Current large language models cannot do any of those things, at least not to the level that uh we'd like them to to do.”He holds that current large language models lack the essential capabilities of intelligence.
2026-09-17 · Sciences Po · 33:37
“ce type de technologie n'est pas un chemin viable vers des systèmes intelligents avec une intelligence similaire à l'intelligence humaine”He believes large language models are not a viable path to human-level intelligence.
2026-08-04 · Zenvora Productions · 56:42
“The main problem we need to solve is how do we learn models of the world? That's and that's what I'm interested. That's what self-supervised learning is all about.”He believes self-supervised learning is the key to building intelligent systems.
2026-07-27 · FRONTia_ch · 01:22
“It's extraordinarily good at processing human language, computer code, and mathematics, but fundamentally unable to apprehend the physical world.”He believes current generative AI is fundamentally unable to understand the physical world.
2026-07-07 · A la french 💻 · 38:12
“le problème du LLM c'est que la seule communication d'information d'une étape de calcul à une autre c'est à Travers un token”He believes LLMs cannot achieve human-like intelligence because they only pass information between computation steps through a few bytes of discrete tokens.
2026-07-07 · A la french 💻 · 34:31
“je ne comprends pas comment les gens peuvent penser s'imaginer qu'ils vont construire des systèmes agentiques qui vont marcher. sans que ces systèmes aient la capacité de prédire les conséquences de leurs actions. c'est complètement débile ça ne peut pas marcher”He thinks agentic systems cannot work without the ability to predict the consequences of actions.
2026-07-07 · A la french 💻 · 24:36
“quand je dis à tout le monde abandonner l'IA générative, je ne me fais pas avec des amis, disons. Mais c'est un objectif, abandonner l'IA générative pour prédire pour des modèles du monde ou une intelligence de type humaine.”He argues that generative AI must be abandoned for world models and human-level intelligence.
2026-07-07 · A la french 💻 · 36:25
“l'intelligence c'est pas une collection de connaissances. ou une accumulation de connaissances déclaratives c'est même pas une collection de compétences sur lesquelles on a appris les recettes c'est plutôt la capacité à inventer de nouvelles recettes”He holds that intelligence is not a collection of knowledge or skills but the ability to invent new solutions when facing new problems.
2026-07-06 · MindVoice Production · 1:06:38
“If you're really interested in human level AI, abandon the idea of generative AI.”He believes generative AI should be abandoned by anyone seriously pursuing human-level AI.
2026-06-30 · MindVoice Production · 09:24
“So, self-supervised learning is, you know, one instance or one attempt at trying to reproduce this kind of learning.”He believes self-supervised learning from observation, not supervised or reinforcement learning, is the key missing paradigm for machine intelligence.
2026-06-30 · MindVoice Production · 1:04:19
“I don't think we can train a machine to be intelligent purely from text. Because I think the amount of information about the world that's contained in text is tiny compared to what we need to know.”He believes machines cannot become intelligent purely from text and must learn grounded world models from sensory data like vision.
2026-06-30 · MindVoice Production · 1:14:41
“I think that's what you know, the entire community should be focusing on at least people who are interested in sort of fundamental questions or you know, really kind of pushing the envelope of AI towards the next the next stage.”He believes the entire research community should focus on self-supervised representation learning and predictive world models as the fundamental challenge.
Blumify can make mistakes. Please check important info.