Yann LeCun’s Post

The narrative of "DeepSeek is the Sputnik moment for US AI" or "the US lost its dominance in AI" seems bizarre to me. DeepMind was founded in London, notable progress on deep learning at University of Toronto, most GPUs made in Taiwan due to TSMC process innovation. Our past progress is due to sharing innovations, of course the same is true for the future.

Haha, indeed. A few years ago, a Silicon Valley CEO asked me: "Why do you even exist?!", referring to that we're competing with Nvidia and Google. His company doesn't exist today. We're prospering and we haven't lost a single customer to neither Nvidia nor Google. BigTech do marvelous things, also in scientific research. The Silicon Valley universities, such as Berkeley and Stanford are of course incredibly successful and have amazing scientists and students. Nevertheless, this culture of brilliance and success also has many blind spots and completely miss out on some of the scientific literature, and what's going on in the rest of the world. Sometimes, this strikes back, as we have seen in the past weeks. It won't be the last time.

This is a great reminder Yann LeCun. Civilizations get lost because of this attitude. Naeem Zafar reminds us of this! Silicon Valley was not like this 20 or 30 years ago. This is a new mentality. Before starting Ascendo AI, I purposefully visited over 20 mid western states and spent time with several entrepreneurs and senior leaders. Many were gracious to spend time and take me out for lunch even though I just knocked on their door! Most don't think about VC. They think about the customer. Several have built companies that have existed 100 years! They are obsessed about product, innovation but deeply care about growth within their teams. This is not just a US-China story. We can learn just within the country.

Couldn’t agree more Yann LeCun! This mindset isn’t just a Silicon Valley issue—it’s across industries. The U.S. has undoubtedly led innovation in many areas, but there’s so much to learn from other parts of the world. Some of the fastest, most resourceful breakthroughs come from places with fewer resources. Look at mobile payments—Africa’s M-Pesa scaled before the West caught on. India’s UPI revolutionized digital transactions, making instant payments accessible to billions. Even in AI, models like DeepSeek are proving that innovation isn’t confined to one geography. Innovation thrives when we learn, adapt, and build on global progress—not dismiss it. The real edge isn’t just leading but staying open to what the rest of the world is doing.

Yann LeCun the syndrome of superiority of good ideas in the US (mainly West Coast) has its based on the venture ecosystem. Good ideas and engineers exist in many places, but good luck achieving anything in your home. Why? Because meritocracy (mostly) exists in the US, anywhere else it's corruption and dogmatism.

Reading between the lines: when he says OpenAI is at the terminal stage of the disease, he is indirectly hinting OpenAI will be dead shortly or become irrelevant.     To be fair, OpenAI had humble beginnings and was an underdog. It was their vision and the ideology that attracted the best talent. But when they had an unexpected success, they could not help keep themselves grounded. And then lost it during the later episodes: boardroom fumbles, internal coups, declaring as a for profit company, going to India and saying "it's hopeless competing with us". All these episodes showed more than a tinge of arrogance.     And they say, "From kings to commoners, history whispers the same truth: arrogance is the prelude to downfall." 

I do not think it is superiority complex - but I hope you are including yourself in that crowd as you lead Meta's AI. Given the layers and complexity of the AI stack from TensorCores to models in Python, there is a whole lot of fluff going around. There has also been lot more focus on alignment, security, explainability etc. rather than de-fating the stack. The narrative of requiring huge capex to train was expected to be a psychological deterrent for anyone thinking of replicating the Ai pioneers. Also, it is surprising that OpenAI was not monitoring their API usage closely. On hind sight, the scaling laws paper became the coda to raise lots of money. That was an empirical study without any rigorous reasoning behind it.

Stage 1: Small tribe syndrome. Stage 2: 'Our ideas are the best!' Stage 3: 'They must be cheating!' Stage 4: ‘Wait, they might have a point...’ Stage 5: ‘Okay, fine, let’s collaborate.’ It's the tech version of the five stages of grief. Today I got to learn that: Silicon Valley's biggest flaw isn't the lack of ideas—it's the ego that comes with believing 'we' own all the good ones. True innovation thrives in collaboration, not isolation. The real magic happens when we build on each other's work, not lock it away. Time to rethink those 'exclusive club' ideologies! Key Takeaways: Success in tech is the product of new ideas, new perspectives, and a more connected world. Tech thrives when ideas flow freely, across borders and beyond egos. Yann LeCun Thank you for sharing your thoughts.

Also work and innovation is not a linear upward trend we often go do unproductive path or take a step backward.

Another symptom - clinging to premises disproved by real-world evidence. The current AI chaos is due to 70 years of narrow and unexamined premises. Key examples: -Premise: language is the source of intelligence -Reality: intelligence is the source of language - babies are intelligent before they can speak -Premise: the brain constructs objects by identifying frequent patterns of object bits -Reality: the brain contains networks of abstract concept cells which fire when the concepts are presented in diverse versions and formats (text, image, sound, etc.) -Premise: neural networks imitate how the brain reasons by processing massive data exhaustively -Reality: the brain computes the relevance of inputs using inhibition networks, filtering out over 99.99%