Meta's chief AI scientist says ChatGPT is 'not revolutionary'

Meta's chief AI scientist, Yann LeCun, is not having the ChatGPT hype.

While the internet treats OpenAI’s viral bot like an alien artifact, LeCun threw cold water on the excitement during a Zoom session hosted by Collective[i] last week. When asked by The New York Times whether Meta’s AI lab, FAIR, would ever achieve a similar breakthrough in the public mind, LeCun bluntly called ChatGPT "not particularly innovative" and "nothing revolutionary."

Great engineering, not rocket science

According to LeCun, the widespread belief that OpenAI stands alone at the frontier of AI is simply inaccurate. Google, Meta, and roughly half a dozen startups possess very similar underlying tech, and there is no secret sauce involved.

LeCun broke down the genealogy of the chatbot to prove his point. ChatGPT runs on self-supervised Transformer architectures. Self-supervised learning is something LeCun pushed long before OpenAI existed, while Transformers were invented by Google in 2017. Tracing the timeline back two decades, Yoshua Bengio built the early neural language models and pioneered the attention concepts that later powered modern architectures.

Even reinforcement learning through human feedback (RLHF), the secret weapon that aligns ChatGPT’s conversational tone, was pioneered by Google’s DeepMind rather than OpenAI. LeCun compared the tool to IBM’s Watson in 2011 or Sebastian Thrun’s DARPA-winning autonomous vehicle in 2005: not a scientific breakthrough, but a piece of solid, well-crafted engineering.

So why didn't Meta or Google ship a chatbot first? LeCun laughed and gave a candid reason: giant tech companies have "a lot to lose by putting out systems that make stuff up."

Still, LeCun promised Meta will roll out comparable tools, focusing heavily on creation aids and generative art. He pointed to the roughly 12 million small mom-and-pop businesses advertising on Facebook that cannot afford creative agencies, arguing AI-generated media could become an essential marketing lifeline for them.

Why it matters

From a pure research standpoint, LeCun is technically correct. The theoretical pillars underpinning ChatGPT belong to Google, DeepMind, Bengio, and decades of shared academic open-source work.

However, we think downplaying OpenAI's achievement as mere engineering misses the entire forest for the trees. Turning fractured academic papers into an intuitive, accessible product that hundreds of millions of people actually want to use is arguably harder than writing another conference paper. OpenAI took the regulatory and reputational leap of faith that incumbents were paralyzed to take.

For builders and creators, the lesson is clear: scientific purity matters far less than product velocity and distribution. LeCun’s dismissal actually highlights the innovator’s dilemma playing out in real time. Meta and Google were sitting on equivalent tech, terrified of hallucinations and public backlash, allowing an agile startup funded by Microsoft to steal the narrative. The AI race is no longer won in arXiv preprints; it is being won in the browser window.