Meta’s CTO, Andrew Bosworth, talked about DeepSeek while answering questions from his Instagram followers this week.
JOSH EDELSON/AFP/Getty
Wall Street may have been caught off guard by DeepSeek’s AI launch, but Meta’s CTO says he wasn’t.Andrew Bosworth said that it was “very predictable” and he wrote an email anticipating such a development 6 months ago.He told his Instagram followers that DeepSeek is “both a big deal and also not as big as it’s made out to be.”
The whole DeepSeek kerfuffle that rattled Silicon Valley? Meta’s CTO says he called such an AI development about 6 months ago.
Asked by one of his Instagram followers on Monday about the buzzy Chinese AI lab’s latest launch, Andrew Bosworth said it “a funny one to watch unfold.”
“I actually had an email of me predicting it, that it would come from somewhere, didn’t know it’d be DeepSeek, like 6 months ago,” the chief technology officer said in an Instagram story. “So I think for those of us in the space, it was not as surprising as those out of it.”
Earlier this month, DeepSeek released its flagship open-source AI model, R1, which is said rivals OpenAI’s o1 model but costs a fraction to develop.
“We were tracking DeepSeek when it launched like a month before it then became this major news item,” Bosworth said. “I think it is both a big deal and also not as big as it’s made out to be.”
Meta has also taken the open-source approach to its Llama AI models, including allowing qualifying researchers to access its individual model weights, the numerical parameters that an AI model learns during training.
“It’s a big deal because it’s a great open-source innovation,” Bosworth said of the DeepSeek launch. “And they’ve done some really truly great and novel work in memory architectures for model building. They’ve certainly advanced the state of the art for reasoning models, they’ve probably done a lot of distilling against existing models.”
Bosworth said DeepSeek news was “a great thing but that’s not a world-changing thing.”
Meta didn’t immediately respond to a request for comment.
On the company’s recent earnings call, Meta CEO Mark Zuckerberg also weighed in on DeepSeek, saying it had done “a number of novel things” that Meta was “still digesting.”
“They have advances that we will hope to implement in our systems, and that’s part of the nature of how this works, whether it’s a Chinese competitor or not,” Zuckerberg said.
While DeepSeek’s cost efficiencies have raised questions on Wall Street about Big Tech’s massive investments in AI infrastructure and top-of-the-line Nvidia chips, Zuckerberg said Meta likely won’t change how it is investing in AI as a result. The Meta CEO said he anticipates Meta will spend hundreds of billions on AI infrastructure in the longterm.
“It’s probably too early to really have a strong opinion on what this means for the trajectory around infrastructure and capex and things like that,” Zuckerberg said.
Meta’s chief AI scientist, Yann LeCun, said in a recent post on Threads that DeepSeek’s launch was a sign that “open source models are surpassing proprietary ones.”
OpenAI CEO Sam Altman, when asked about DeepSeek and the open-source practice of releasing AI model weights and publishing research, said he was considering making some changes.
“Yes, we are discussing,” Altman said. “I personally think we have been on the wrong side of history here and need to figure out a different open source strategy; not everyone at OpenAI shares this view, and it’s also not our current highest priority.”
Do you work at Meta? Reach out to the reporter from a non-work email and device at sjackson@businessinsider.com
+ There are no comments
Add yours