What Yann LeCun said
Written from the verified transcript and checked against it. Every figure links to the moment it was said.
Yann LeCun argued that Japan's lack of confidence in its manufacturing and data assets is limiting its AI implementation, urging the country to leverage its world-class precision manufacturing and high-quality data to become a first mover in physical AI. He introduced Project Tapestry, an open, collaborative initiative to build a global AI model where countries contribute computing resources and training data while preserving data sovereignty. He criticized current LLM-based AI for being unable to handle high-dimensional continuous data, predict consequences, or achieve human-level intelligence, and instead advocated for world models and JEPA (Joint Embedding Predictive Architecture) as the blueprint for safe, agentic physical AI. LeCun encouraged students to abandon generative models and work on physical AI, which he sees as the next revolution.
Key takeaways
Japan should have greater confidence in its physical industry and data advantages to lead in physical AI. Project Tapestry enables countries to contribute to a global open AI model while retaining data sovereignty. LLMs cannot achieve human-level intelligence because they lack world model prediction and sensor data handling. JEPA architecture abandons generative models to predict in abstract representation space, enabling safe planning. Students should focus on physical AI, not LLMs, to bring AI to human level.
Numbers and commitments
Chapters
0:00 Japan's lack of confidence 1:44 Japan's advantage in data 2:33 Project Tapestry and open AI 3:46 AI sovereignty 4:56 Physical AI as next revolution 5:56 Limitations of LLMs 6:48 Intelligence and world models 11:15 JEPA architecture and safe AI
▲
▼
EXEC ONLY
Forget about your native AI.
Where it's more than just your data, it's your identity, it's your knowledge, it's yourself.
Cite this transcript APA, MLA, BibTeX APA Copy
LeCun, Y. (2026, October 27). 【AIの父 ヤン・ルカン】「LLMに取り組むべきではない」ワケ／ChatGPTでは人間レベルに届かない？／10歳児にできてAIにできないこと／次世代AI「世界モデル」の正体 [Interview transcript]. ビジネス＋ITチャンネル. CEOInterviews.AI. https://ceointerviews.ai/interview/3147399/
MLA Copy
Yann LeCun. "【AIの父 ヤン・ルカン】「LLMに取り組むべきではない」ワケ／ChatGPTでは人間レベルに届かない？／10歳児にできてAIにできないこと／次世代AI「世界モデル」の正体." ビジネス＋ITチャンネル, 27 Oct. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/3147399/.
BibTeX Copy
@misc{lecun2026_3147399,
author = {Yann LeCun},
title = {【AIの父 ヤン・ルカン】「LLMに取り組むべきではない」ワケ／ChatGPTでは人間レベルに届かない？／10歳児にできてAIにできないこと／次世代AI「世界モデル」の正体},
howpublished = {Interview transcript, ビジネス＋ITチャンネル. CEOInterviews.AI},
year = {2026},
month = {oct},
url = {https://ceointerviews.ai/interview/3147399/},
note = {Speaker-attributed transcript with timestamps}
}
▶
Key Quotes (2)
“So, currently AI sucks, if you pardon my French. When we compare the learning abilities of AI systems today with the learning abilities of animals and humans, machines learn extremely slowly. It turns out manipulating discrete symbols like language or code or mathematics is much easier than understanding the real world. The main limitation, the main reason for this is that current approaches to AI”
“We're not going to get to human-level intelligence by just training on text. Now, we learn a lot more information about how the world works through our senses, vision, touch, audition, than we ever can through text. And that means we're never going to get to human-level intelligence or physical AI by training on text. So, this is where world models come into play. The way a system of the type I de”
Mentioned in this interview
Jean Piaget : Quoted his definition of intelligence: 'what you do when you don't know'
Earlier, on the same topics
“Generative approaches should be thrown away if what you are interested in is AI that understands the real world — they work for language because language is discrete, but they fail for continuous sensory data. Instead we should build joint-embedding predictive architectures (world models that predic”
“We need a high diversity of AI assistance for the same reason we need high diversity of the press. Because they're going to mediate all over information. Now, this is not going to happen with the current system. What is required for this is AI what we call AI foundation models that are open, that ar”
“The long term is: where is my domestic robot that can clear the table, load the dishwasher, clean the bathroom, do the housework? Today it can talk to me, roughly. Today it can talk but it can't do anything, or it can do extremely narrow tasks for which it was trained with enormous amounts of data b”
“I do not understand how you can even think of building an agentic system without a agentic system having the ability of predicting the consequences of its actions. Okay. And VLA doesn't doesn't do that. Sure. Right. VLAs do not have world models. They cannot predict the consequences of their actions”
Every Yann LeCun interview →
Newsletters that week
Beefcake! Exploring ChatGPT · Exploring ChatGPT · Nov 02, 2026
More interviews with Yann LeCun
AI at 70 Award Presentation & Keynote: World Models: Enabling the Next AI Revolution Oct 29, 2026 · Dartmouth · 64 min Yann LeCun, AMI Labs | Evident AI Symposium 2026 Oct 27, 2026 · Evident Insights · 19 min 【サイバー攻撃相次ぐ今は「移行期に過ぎない」ヤン・ルカン】公開前にAIで自社ソフトを攻撃→弱点を補強／GPT暴走事故は「人的ミス」／最先端モデル「秘密にすれば安全」は間違い【1on1 Tech】 Oct 27, 2026 · TBS CROSS DIG with Bloomberg · 44 min Yann LeCun, AI legend, Co-Founder of AMI Labs, NYU Professor [AI Founders Masterclass Session 5] Oct 22, 2026 · Radical Ventures · 57 min [Long version] World Models, a game changer? With Yann LeCun | AIM 2026 Oct 01, 2026 · La Tribune Événements · 49 min Yann LeCun: Where Is Artificial Intelligence Heading? Sep 16, 2026 · Sciences Po · 88 min
All Yann LeCun transcripts →
About Yann LeCun
Yann LeCun, co-founder and executive chairman of AMI Labs, has continued to publicly criticize large language models (LLMs) as an insufficient path to human-level AI. In appearances throughout late 2026, he described LLMs as "rote learners" and said that "true progress in AI will come because of new concepts," arguing that current systems lack the ability to reason, plan, or understand the physical world. He has advocated for "world models" and his JEPA architecture, which predicts in an abstract representation space rather than reconstructing every detail, as a more promising approach. LeCun also expressed skepticism about existential-risk claims, stating there is "no science behind it at all," and rejected the term "AGI" as "complete nonsense," preferring "human-level AI" or "advanced machine intelligence."
LeCun has promoted open-source AI, saying that diverse AI platforms are needed to avoid control by a small number of companies and that open models are more secure because "there's more eyeballs on it." He has also been active in conference keynotes and interviews, including receiving the AI at 70 Award at Dartmouth, discussing Japan's manufacturing strengths in physical AI, and presenting his research at industry events. In these engagements, he repeated his view that training on text alone will not achieve human-level intelligence and that future progress likely depends on physical AI and robotics, though he acknowledged current humanoid robot companies lack a clear path to making their products "smart enough to be useful."
All Yann LeCun appearances →
ビジネス＋ITチャンネル on YouTube →
