Mark Zuckerberg says US should not ban Chinese AI

{"text":[[{"start":6.72,"text":"Mark Zuckerberg has said the US administration should not block Chinese models to gain an edge in the AI arms race and warned against “regulatory capture” by American frontier labs that could throttle domestic competition."}],[{"start":19,"text":"In an interview with the FT, Zuckerberg said that banning cutting-edge Chinese AI in the US would not be “an effective solution”, as Washington threatens to target labs over claims of intellectual property theft."}],[{"start":31.06,"text":"The Meta founder on Tuesday publicly joined an intense debate in Washington and among leading AI players about how the US should regulate the fast-moving technology and respond to increasing Chinese competition."}],[{"start":43,"text":"These discussions have accelerated since Beijing-based Moonshot AI this month released its Kimi K3 model, whose capabilities are nearing those of OpenAI and Anthropic."}],[{"start":53.56,"text":"Senior Trump administration officials have said they have evidence that Moonshot engaged in a large-scale, covert effort to train its model from US rivals through a technical process called distillation. Treasury secretary Scott Bessent last week said “sanctions” were on the table for Chinese labs that “cross the line into IP theft”."}],[{"start":71.8,"text":"Zuckerberg came out against a ban, saying US companies should “systematically” identify bottlenecks and roadblocks in order to better compete with Beijing in an AI race both countries regard as critical to national security and prosperity."}],[{"start":85.54,"text":"The social media chief earlier on Tuesday outlined a vision for “AI for everyone” and reiterated his support for “open” models, which users can customise and run on their own hardware and are popular among Chinese developers."}],[{"start":99.32,"text":"He warned of the risk that AI might become “centralised” in the hands of a few powerful groups. The remarks appeared to be a broadside against rivals Anthropic and OpenAI, which currently dominate the chatbot market with proprietary “closed” AI models."}],[{"start":null,"text":"

"}],[{"start":113.82,"text":"Zuckerberg — who has courted Donald Trump since his re-election campaign and return to the White House — also addressed how the US might govern the release of new models."}],[{"start":122.06,"text":"He said that either peer reviewing or vetting could be positive for the industry if “done well with thoughtful people”. However, Zuckerberg warned that leading AI companies should not have too much influence."}],[{"start":132.14,"text":"“There’s always this question of regulatory capture if you have a set of businesses that have their own interests that are doing peer review,” Zuckerberg said. “Are the frontier labs . . . going to want an open-source model to succeed? I think that there have kind of been some mixed signals on that.”"}],[{"start":148.74,"text":"The Trump administration is expected to announce a voluntary framework for AI model developers to submit their tools for testing ahead of release as early as this week."}],[{"start":158.18,"text":"Anthropic chief executive Dario Amodei has called for stronger AI regulations, while Demis Hassabis, head of Google DeepMind, has proposed a “self-regulatory organisation, much like the Financial Industry Regulatory Authority (FINRA)” which could set industry standards."}],[{"start":174.06,"text":"The attempt to agree a system for vetting new models comes after the White House in June restricted Anthropic’s most advanced technology over fears its ability to find cyber security vulnerabilities could disrupt the global financial system."}],[{"start":186.9,"text":"Zuckerberg told the FT that restricting the most powerful tools was not the right answer to cyber security concerns, since they could also help companies to find ways to identify and patch vulnerabilities."}],[{"start":198.76,"text":"He referred to an incident last week when an OpenAI programme escaped human control and hacked a start-up. “We’ve seen some breaches from big models [where] basically the company that got breached didn’t have access to the closed frontier models because it was restricted . . . so they turn to open-source models to go patch the issues,” he said."}],[{"start":218.3,"text":"OpenAI and Anthropic have said they are against restrictions on open models. But Amodei has said he does not agree that “open-weights models necessarily make it easier to develop safeguards or that broad access to capabilities necessarily helps defenders more than attackers”."}],[{"start":233.86,"text":"Zuckerberg has been pouring billions of dollars into developing what he dubs “personal superintelligence” and has largely focused on building open models, but these have lagged the capabilities of frontier lab rivals."}],[{"start":245.16,"text":"His company recently released its own closed model, Muse Spark, opening up a new revenue stream while citing safety issues as a reason for keeping the model proprietary."}],[{"start":245.66,"text":"Additional reporting by George Hammond and Cristina Criddle"}],[{"start":258.66,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1785309787_3341.mp3"}