Geoffrey Hinton vs Yann LeCun: Two AI Pioneers on Risk, Open Models, and What Comes Next
Comparing the publicly stated views of Geoffrey Hinton and Yann LeCun on AI risk, regulation, open models, and the path beyond large language models: what each has said, and where they differ.
Geoffrey Hinton and Yann LeCun are two of the three researchers often called the godfathers of deep learning. Since 2023 they have become public voices on opposite sides of the debate over how dangerous advanced AI might be. People compare them to understand the disagreement. This guide reports their documented positions and does not take a side.
What is the core difference?
Hinton has said he became more concerned about the capabilities of large neural networks and about the possibility that AI systems could outthink humans and pursue goals that conflict with ours. He emphasizes the need for research and regulation.
LeCun argues that current systems are far from human-level intelligence, that they lack key abilities, and that the scenarios in which AI takes over are speculative. He emphasizes open development and the benefits of the technology.
One urges caution about what might come. The other emphasizes what current systems cannot do and the value of openness.
What has Hinton said?
Geoffrey Hinton, a British-Canadian computer scientist, co-authored the 1986 paper on backpropagation with David Rumelhart and Ronald Williams and led work on deep neural networks for decades. His students' 2012 AlexNet result helped start the modern deep learning boom. He worked at Google until May 2023, when he resigned in order to talk about AI risks. He shared the 2024 Nobel Prize in Physics with John Hopfield.
Since leaving, he has described risks from misuse, job disruption, and the possibility that AI systems may become more intelligent than humans, and has called for safety research and regulation. His statements are in interviews, talks, and public letters.
What has LeCun said?
Yann LeCun, a French-American computer scientist, developed convolutional neural networks for image recognition in the 1980s and 1990s, and later led AI research at Meta. He is a professor at New York University. He has argued that current large language models are limited, lacking a model of the world, persistent memory, and planning, and that further progress will require new architectures.
He opposes proposals to restrict open-source AI, arguing that openness enables scrutiny and competition. His views are expressed in talks, interviews, social media, and in a 2022 paper outlining a path to autonomous machine intelligence. His positions are widely debated.
How do they compare?
| Dimension | LeCun | Hinton |
|---|---|---|
| Public stance on AI risk | Existential-risk fears overstated | Serious risks that warrant caution and research |
| View of current systems | Limited; lack world models and planning | Capable enough to take seriously, with fast progress |
| Open models | Strongly in favor | Concerned about misuse of powerful models |
| Regulation | Skeptical of heavy restrictions on open research | Supports safety research and regulation |
| Background | Convolutional networks; Meta; NYU | Backpropagation, deep learning; Google; Toronto |
| Shared | 2018 Turing Award with Hinton and Bengio | 2018 Turing Award; 2024 Nobel Prize in Physics |
| Best sources | Talks, posts, and technical papers | Interviews, talks, and academic papers |
When does each view help?
Reading Hinton helps if you want to understand the case for taking AI risks seriously from a pioneer of the field, including his reasons for changing his view. It frames questions about safety and governance.
Reading LeCun helps if you want to understand the technical case that current systems fall short of general intelligence, and the case for openness. It frames questions about what capabilities are missing.
Neither settles the matter. Readers who want to weigh the debate also read other experts, including those who work on AI safety, policy, and alternative technical paths.
What does this look like in practice?
A company is deciding whether to release an AI model openly. A Hinton-informed view would ask what misuse the model enables, how capable future versions might be, and what safeguards are possible before release. It leans toward caution and evaluation.
A LeCun-informed view would ask what benefits openness brings, such as scrutiny, innovation, and competition, and whether the model's capabilities justify the feared risks. It leans toward release with attention to realistic harms. A sound decision examines specific capabilities and harms, and not only the general stance.
"Summarize the strongest version of Hinton's public case about AI risk and the strongest version of LeCun's public case that such fears are overstated. For each, list the evidence offered, the assumptions it depends on, and what future developments would count against it. Then list the questions that experts in between the two still debate. Do not tell me who is right."
Why it works: separating evidence from assumptions lets you evaluate the debate without picking a camp first.
Can you use both together?
As sources, yes. Reading both exposes the strongest versions of the opposing views, and the points on which they agree: that the technology is powerful and that careful thought is warranted.
Be wary of tribal framing. The field includes many views, and individual claims should be judged on their arguments.
"For [AI system or policy I am considering], translate the general risk debate into specific questions: what capabilities does it have, what misuse is plausible, what safeguards exist, and what benefits would be lost by restriction? Note which questions Hinton's concerns would raise and which LeCun's arguments would answer."
Why it works: specific capabilities and harms are easier to assess than general positions on the future of AI.
Where to go next
Geoffrey Hinton and Yann LeCun both sit in the AI Researcher category. For primary sources, see Hinton's interviews and academic papers and LeCun's talks, posts, and 2022 position paper. This guide reports public positions and takes no side. To work through a major decision in a structured way, Decision Brief is a $79 tool built for it.
Frequently asked questions
Are Hinton and LeCun opponents?
They are long-time colleagues who share credit for foundational work on deep learning and, with Yoshua Bengio, received the 2018 Turing Award. Since 2023 they have disagreed publicly on how serious the risks of advanced AI are and how to regulate it. The disagreement is about risk assessment and policy, not about the value of the field they built together.
What has Hinton said about AI risk?
In May 2023 he left Google in order to speak freely about the dangers of AI, saying that systems might become more intelligent than humans and that this could pose serious risks, including misuse and loss of control. He has called for research on safety and for regulation. These are his public statements, and other experts, including LeCun, disagree with parts of them.
What has LeCun said?
LeCun has argued publicly that fears of existential risk from AI are overstated, that current large language models lack key capabilities such as planning and a model of the world, and that open research and open-source models are beneficial and safer than concentrating control in a few firms. He has proposed alternative architectures, such as joint embedding predictive architectures, as steps toward more capable systems.
Who is right?
The question is contested and this guide does not settle it. Both are leading experts, and the future capabilities and risks of AI are uncertain. Readers should look at the arguments, the evidence offered, and the views of other researchers, including those who sit between the two. This comparison reports public positions and does not endorse either. Instead, look at what each predicts, what would show them wrong, and what other researchers say about those predictions.
Where can I read their views directly?
Hinton has given many interviews and talks since 2023, and has a long record of academic papers. LeCun posts regularly on social media and has given talks and interviews, and has published technical papers, including a 2022 position paper on a path toward autonomous machine intelligence. Primary sources give a clearer picture than summaries, including this one. Reading their words in context avoids the distortions of headlines and short clips.
Can AI help me study this debate?
It can summarize arguments and list the points each side makes, but it may reproduce errors and reflect its own training. Verify quotations and dates against primary sources, and treat it as a way to organize the debate rather than a judge of it. Verify quotations and dates, and keep in mind that views in this field change quickly. Cross-check them with more than one source.
Written by Gareth Hoyle. Last updated 8 October 2026. Part of the authority.md guides library.
More guides.
Daron Acemoglu vs Ha-Joon Chang: Do Good Institutions Cause Growth, or Follow It?
Comparing Daron Acemoglu's institutions-first account of national prosperity in Why Nations Fail with Ha-Joon Chang's historical critique of the institutions literature: two views of development, with the main point of disagreement. General education.
Bill Ackman vs Carl Icahn: Two Styles of Activist Investing. Not Investment Advice
Comparing Bill Ackman's public, thesis-driven activism with Carl Icahn's decades of corporate pressure campaigns: how each approaches companies, boards, and the public. Not investment advice.
Adam Grant vs Simon Sinek: Evidence and Rethinking or Purpose and Why
Comparing Adam Grant's research-based work on giving, originality, and rethinking with Simon Sinek's purpose-driven Start With Why and Infinite Game: two popular voices on leadership, with different relationships to evidence.