What happens when one of AI's godfathers says the whole industry is building the wrong thing? скачать в хорошем качестве

What happens when one of AI's godfathers says the whole industry is building the wrong thing?

What happens when one of AI's godfathers says the whole industry is building the wrong thing? Yann LeCun (Turing Award winner, inventor of convolutional neural networks, the guy who literally built a huge part of the foundations of modern AI) is no longer with Meta. No, he didn't retire. He launched AMI Labs in Paris to build what he calls "world models." At India's AI Summit recently (the one where Sam and Daario wouldn’t hold hands) Yann LeCun told every academic in the room: "Don't work on LLMs. There is no point. You're not going to rival what's going on in industry." He compared LLMs to the printing press: transformative, sure, but fundamentally not intelligence. LeCun's argument isn't that LLMs are useless. It's that we've confused a very sophisticated information retrieval system with actual reasoning. His test case: a seventeen-year-old can learn to drive a car in twenty hours. No LLM can do that. Not because we need more data or bigger models. Because the architecture is fundamentally wrong for understanding how the physical world works. (And the seventeen-year-old, notably,can learn to drivewhile texting.) OpenAI, Anthropic, Google... they're all sprinting to make LLMs bigger. LeCun's betting the real breakthrough comes from something none of them are building yet. He's either brilliantly early or spectacularly wrong. Either way, it's worth paying attention when a godfather walks away from the family business. If LLM's aren't the way, how do we achieve AGI?