The AI revolution is well and truly underway, but OpenAI CEO Sam Altman believes it could’ve been set off even sooner had it not been for the experts in the field.
In a recent interview, Altman was candid about what he sees as a systemic failure of scientific judgment — one that delayed progress by years. “The field was honestly held back by a generation of scientists who just were way too certain on what scaling was not going to produce results,” he said. “And then some people just looked at the graphs and said, well, it looks like it’s continuing beautifully, let’s keep going.”
He did acknowledge that “world models are clearly important and we’ll need that for things like robotics” — a concession that reads almost as a nod toward Yann LeCun, Meta’s former Chief AI Scientist and the field’s most prominent LLM skeptic. LeCun has spent years arguing that LLMs are fundamentally limited — that they are “just token generators,” that text data alone is insufficient for human-level AI, and that true intelligence requires understanding the physical world. He has since left Meta to found AMI Labs, a company explicitly built around non-LLM world models, backed by a $1.03 billion seed round. But Altman was unambiguous about where he stands: “Betting against LLM scaling at this point feels quite misguided to me.”
When the interviewer asked whether it gets annoying to be the “I-told-you-so guy,” Altman’s answer was revealing. “There are these Twitter trolls that for years have just been like, it’s not going to work, it’s not going to work, this is so dumb, this is a fraud, this company is going to fail, this research approach is going to fail,” he said. “I used to get more bothered by them, but I don’t even feel the I-told-you-so at this point. You were just there. You’re still going on about it. The data is quite strong on our side.”
What’s interesting is the particular way Altman chose to frame the holdouts. He didn’t describe them as simply wrong — he described the behaviour itself as a kind of pathology. “There’s that saying that insanity is doing the same thing over and over again when presented with data that is not working and they keep repeating that,” he said. “In a sense, it’s a form of insanity.”
He went further, identifying what he thinks drives it. “If you make your identity about a particular thing is going to work or not work, and you associate yourself with that belief, and then the science or the empirical results disprove you, and you’re too hung up on your identity, you can’t let it go, you can’t see the truth,” he said. “And I think this is an important reminder in both directions” — a rare moment of self-awareness in an otherwise pointed critique.
The implied target of much of this is hard to miss. LeCun has been publicly and consistently bearish on LLMs for years, arguing that they can never build genuine world models and that text data alone cannot produce human-level intelligence. At NVIDIA GTC 2025, he declared he was no longer interested in LLMs and pivoted his focus to architectures capable of understanding the physical world, persistent memory, and genuine planning. The positions are well-documented and widely cited, and they represent exactly the kind of identity-anchored scientific bet Altman was describing.
That said, the broader debate is far from a simple case of one side being vindicated. LLMs have scaled impressively, and OpenAI’s commercial success is hard to argue with. But the questions LeCun and others have raised — about reasoning, grounding, and physical understanding — haven’t gone away, and the robotics gap Altman himself acknowledged is real. What Altman seems to be pushing back against is less the substance of those concerns and more the certainty with which critics ruled out an entire paradigm before the results were in. Science, his comments suggest, should follow the graphs.