5 Nov 2026 Yann LeCun has spent years telling the AI industry it’s making a huge mistake, and a new paper might be the strongest evidence yet that LeCun was right. In “Proving and Improving the Platonic Representation Hypothesis with JEPAs”, Randall Balestriero and Yann LeCun show that a 109 million parameter JEPA beat language models up to 72 billion parameters on how well it lines up with the way image models see the world, with about 42x less training data. In this video I explain what LeCun actually found, in plain English, and what it could mean if this idea is scaled up.
⏱️ Chapters
- 0:00 Intro
- 1:33 Who Yann LeCun is (and why I keep talking about LeCun)
- 4:25 The mistake LeCun says the AI industry is making
- 8:05 World models and JEPA, explained simply
- 11:56 What the new paper actually found
- 15:54 What the paper doesn’t prove
- 17:56 What happens if this is scaled up
- 20:34 Could this be the path to real AGI?
- 23:03 The other big implications
- 25:12 LeCun is moving faster than people realise
- 27:02 Why this could still break at scale
- 29:01 My honest take
🎥 Video clips courtesy of: Big Technology Podcast, AI House Davos, This Is World, World Science Festival, Unsupervised Learning (Redpoint), ETH Zurich (Computer Vision and Geometry Group), NVIDIA GTC, World Labs, 3D World Simulator team (Columbia, UC Berkeley, University of Michigan, NVIDIA), @slow_developer
📺 My earlier videos on Yann LeCun:
Yann LeCun Just Called Out the Entire Robotics Industry: • Yann LeCun Just Called Out the Entire Robo…
AI Experts Are Warning About a Dangerous New Problem With LLMs: • AI Experts Are Warning About a Dangerous N…
Yann LeCun Goes OFF On Elon Musk’s xAI: • Yann Lecun Goes OFF On Elon Musks XAI "The…
🐦 Posts featured: @ylecun, @Arjunjain, @continuumlabs_, @heyshrutimishra, @houlemarde, @BobClowdr
📰 Sources
- The paper, Proving and Improving the Platonic Representation Hypothesis with JEPAs (arXiv): https://arxiv.org/abs/2610.39302
- The paper on alphaXiv: https://www.alphaxiv.org/abs/2610.39302
- The original Platonic Representation Hypothesis (MIT, 2024): https://arxiv.org/abs/2405.07987
- Paper explainer (36Kr): https://eu.36kr.com/en/p/405208007742…
- Paper explainer (Startup Fortune): https://startupfortune.com/yann-lecun…
- Paper explainer (Emergent): https://emergent.sh/news/lecun-paper-…
- LeJEPA (Nov 2025): https://arxiv.org/abs/2511.08544
- AMI Labs raises $1.03B (TechCrunch): https://techcrunch.com/2026/03/09/yan…
- AMI Labs roadmap, LeWorldModel and H-JEPA (HyperAI): https://hyper.ai/en/stories/a7b52e687…
- When Does LeJEPA Learn a World Model? (StartupHub): https://www.startuphub.ai/ai-news/ai-…
- LeWorldModel explained: / leworldmodel-how-yann-lecun-solved-the-har…
- Advanced Machine Intelligence Labs: https://en.wikipedia.org/wiki/Advance…
- Yann LeCun: https://en.wikipedia.org/wiki/Yann_LeCun
- Convolutional neural network: https://en.wikipedia.org/wiki/Convolu…
- LeNet: https://en.wikipedia.org/wiki/LeNet
- Turing Award: https://en.wikipedia.org/wiki/Turing_…
- The Bitter Lesson: https://en.wikipedia.org/wiki/Bitter_…
- World models: https://en.wikipedia.org/wiki/World_m…)
- Yann LeCun, World Models (ETH Zurich lecture): • Yann LeCun: World Models: Enabling the nex…
- AI House Davos 2026, Embodied AI: • Embodied AI: Systems that See, Hear, and A…
- Big Technology Podcast, Yann LeCun: • Yann LeCun: We Won’t Reach AGI By Scaling …
- Unsupervised Learning, Yann LeCun on What Comes After LLMs: • Yann LeCun on What Comes After LLMs
- World Labs, Atlas: https://www.worldlabs.ai/blog/atlas
- 3D World Simulator: Hybrid Particle-Voxel Dynamics for Multimodal 3D World Simulation: https://jwang771.github.io/3dworldsim…
#AI #YannLeCun #WorldModels #JEPA #AGI
- channel url: https://www.youtube.com/@TheAiGrid


