Yann LeCun, Meta’s top AI scientist (R), told CEO Mark Zuckerberg that OpenAI’s ChatGPT posed a risk to Instagram and Facebook.
Andrej Sokolow, Getty Images; Kevin Dietsch, Getty Images
Meta’s top AI scientist urged Mark Zuckerberg to make a ChatGPT rival, the New York Times reports.Yann LeCun warned that Facebook and Instagram could potentially go extinct if Meta didn’t act. A few months later, Meta introduced LLaMA, but the AI assistant hit trouble after a coding leak.
Meta’s chief AI scientist warned CEO Mark Zuckerberg that OpenAI’s ChatGPT could spell trouble for the company — and that he needed to act fast before it was too late.
Six weeks after OpenAI launched ChatGPT in November 2022, Yann LeCun got real with Zuckerberg as they waited in line for lunch at Meta’s headquarters in Menlo Park, California, a source familiar with the matter told The New York Times.
The scientist told the CEO that Meta needed to catch up with OpenAI’s technology and make moves on releasing an its own AI assistant, the report said.
If Meta didn’t buckle down on its AI efforts, Facebook and Instagram could go extinct, LeCun warned, according to the report.
Zuckerberg didn’t respond, the Times reported, though it said that LeCun’s comments irked him. After all, Zuckerberg wanted Meta to be known as a leading AI company, and Meta had already implemented AI features across apps like Facebook, Instagram, and WhatsApp.
Still, later in the evening, Zuckerberg appeared to have a change of heart.
“I have been thinking about what you said,” the CEO told LeCun during a dinner that evening, a person familiar told the Times. “And I think you’re right.”
Meta didn’t immediately reply to Business Insider’s request for comment from Zuckerberg or LeCun.
Meanwhile, despite some what the Times reported as pushback from Meta’s legal and policy teams over safety concerns, Zuckerberg decided to move forward with releasing the company’s own open-sourced AI model as quickly as possible.
By February, Meta had renamed Genesis, the large language model LeCun’s team built, to LLaMA, and released it to researchers outside of the company.
But days after the initial release, the model’s code got leaked on 4chan, an online messaging board, playing into internal security worries. The leak partially spurred Senators Josh Hawley of Missouri and Richard Blumenthal of Connecticut to write to Meta in the months following, alleging that the company “failed to conduct any meaningful risk assessment” before releasing LLaMA and made the model vulnerable to bad actors.
Meta wasn’t alone in the AI race, though. Tech rivals like Google and Microsoft were scurrying to launch their own chatbots to beat OpenAI — also with some mixed results.
In February, the same month Meta released LLaMA, Microsoft released its AI-powered Bing search engine to early users, which was later accused of spitting out disturbing responses. That same week, Google unveiled its beta version of its chatbot Bard, which some employees criticized for being a “rushed” and “botched” job after it made a factual error during its first demo.
The experience didn’t put Meta off; it continued its pursuit of new AI tools with its July release of Llama 2, the next iteration of its initial model.
In September, it also unveiled its newest generation of Ray-Ban Meta smart glasses, and dozens of AI chatbots that play the roles of celebrities like Paris Hilton and Snoop Dogg.