Meta Chief AI Scientist Yann LeCun says ChatGPT is “not particularly innovative” in its underlying techniques and other research labs are doing similar work
Much ink has been spilled of late about the tremendous promise of OpenAI's ChatGPT program for generating natural-language utterances in response to human prompts.
ZDNetTiernan Ray
Context & Ripple Effects
LeCun's assessment lands as generative ML was being treated as a step change in practical AI use cases, while OpenAI's release had already triggered Google's internal search response. The dispute is therefore less about whether the techniques are novel than about who turns them into broadly used products.
For Meta, the contrast is acute: related coverage describes an AI leader that was nonetheless cautious with public chatbots after closing Galactica over toxic content and misinformation. LeCun's comments defend research parity while Meta confronts the product-safety trade-offs of deploying comparable systems.
First-order effects
Meta gains a public argument that OpenAI's chatbot lead does not establish a unique technical moat, even as OpenAI retains the attention generated by ChatGPT's release.
OpenAI faces more scrutiny of its product execution—training, safety, and user experience—rather than being judged solely on the novelty of its underlying methods.
Second-order effects
Google's code-red response shows that rivals must respond to ChatGPT's user adoption even if they agree the core techniques are reproducible; distribution and integration become the immediate competitive variables.
Meta's Galactica experience raises the cost of a fast follower strategy: safety and misinformation controls can determine whether research capability becomes a public chatbot.
Third-order effects
If leading labs continue to share broadly similar generative techniques, durable advantage is likely to shift toward product distribution, trust controls, and integration into existing services rather than model novelty alone.
The pattern points to chatbot competition becoming an assistant-layer contest, with public reliability and deployment governance separating labs that can ship from those that primarily demonstrate research.
The trend:Generative AI is moving from a debate over model breakthroughs toward competition over safe deployment, distribution, and product integration.
The title is blunt, but the article conveys what I said on the @collectivei Q&A about progress in AI. ChatGPT & other LLMs didn't come out of a vacuum & are the results of decades of contributions from various people. No AI lab is significantly ahead of the others. https://twitte…
Even OpenAI folks have been like “Can't believe no one made this first”, since the approach and scale aren't terribly new. On the other hand, it is definitely revolutionary because normal people are using it and enjoying it, in a way unlike LLMs before it. https://twitter.com/...
ChatGPT is ‘not particularly innovative,’ and ‘nothing revolutionary’, says Meta's chief AI scientist https://www.zdnet.com/... by @TiernanRayTech via @ZDNET
Stephen Wolfram had a great observation about this: If a tool or idea is truly paradigm-changing, then it will undermine people's life's work and you should expect a backlash. There's a reason so many AI researchers dislike ChatGPT. https://twitter.com/... https://twitter.com/...
You know someone's lost the plot when they say things like “it's nothing revolutionary” to a product that's caught lightning in a bottle. That's not the point. The point is that tons of people find it actually useful. Shame to see. https://www.zdnet.com/...
@nathanbenaich Sounded like OpenAI weren't aware how popular it would actually be. Maybe technically ChatGPT isn't that revolutionary, but it definitely crossed some uncanny valley threshold of usability.
@ylecun @collectivei Some labs seem to be farther ahead in public relations though... It seems the perception of OpenAI being ahead is pretty significant.
“In terms of underlying techniques, ChatGPT is not particularly innovative,” @ylecun shared during last week's #ciFORECAST. Check out the below article for an overview of Yann's predictions and insights. https://www.zdnet.com/... #artificalintelligence
I am an ML researcher but I can say: the interface, product design, amount of ole' RLHF — something is very different than other LLMs because the social impact is drastically bigger.
“In terms of underlying techniques, ChatGPT is not particularly innovative or revolutionary, although that's the way it's perceived in the public. It's just well put together.” -Yann LeCun, Meta's chief AI scientist https://www.zdnet.com/... @ZDNET @TiernanRayTech
@ylecun I do agree with your initial tweet, LLMs “do make sh*t up”. Why do you think this would be fine for a writing assistant like Galactica? Making up papers that don't exist isn't particularly helpful, indeed, many would (and have) argue it is harmful. Curious on your take.
This maybe true but is the race now only about technology? They have a giant distribution partner and are out selling, winning customers and learning #ChatGPT https://twitter.com/...