Zuckerberg Reflects on Meta AI: 'Llama 4 Forced Me to Shift My Thousand-Person Team into Small-Scale Research Groups—A Year Later, This Decision Gave Rise to Muse'

The host asked him: 'What does building things feel like for you?' His answer came instantly: 'I just need to build stuff. I just think if I'm not exercising my creativity and building stuff I get grumpy.' He compared this drive to others’ motivations: some people need writing to express themselves; some need to be liked; he needs to build. Without building, he becomes irritable—and those around him prefer he *not* be irritable. He frames building as a fundamental human need, akin to breathing—deeper than merely saying 'I love entrepreneurship.' Then he recounted a scene from Priscilla’s first day of medical school. Someone asked the students: 'How many of you remember, as a child, seeing someone and thinking, "I really want to take care of that person"?' Everyone raised their hand. Zuckerberg said his version of that memory was: 'I want to build this thing—and make it better.' A doctor’s drive is caregiving; his is building. Same structure—different direction. On learning, his view is refreshingly simple. He studied Mandarin for over a year, learned martial arts, and even trained to fly helicopters. He believes these share one key trait: they can’t be mastered through pure reasoning alone. You simply have to sit down, invest time, and let knowledge gradually seep into your brain. Programming has a theoretical component—but product intuition comes only from hands-on practice. He extended this idea to teaching children. One of his daughters loves making music but hates piano lessons. He told her: 'You don’t need to become a piano master—but if you want to compose music, you need an intuitive grasp of music theory. Once you learn piano, picking up guitar becomes easy.' The core idea: identify what truly matters to you—and then practice with discipline to build instinctive fluency. 11. First write 15 pages to clarify your thinking—then distill it layer by layer into a single short video. This summer, Zuckerberg published a 6,500-word essay titled 'The Future Is for Everyone,' systematically laying out his stance on AI-related societal issues: What role should governments play? How do we protect national security? How should data centers coexist with local communities? How can AI create jobs instead of eliminating them? How do we guard against cyberattacks and biosecurity risks? These topics are highly complex. He consulted extensively, revised multiple drafts, and engaged in intense internal team debates—eventually producing a 15-page document. This writing phase wasn’t intended for public release; its purpose was to help *him* think clearly. Once completed, those 15 pages became the vehicle for internal alignment: 'This is what we believe.' Then he began cutting back. The 15-page document was condensed into a 1-page op-ed—stripping away all supporting arguments and case studies, retaining only core judgments. That op-ed was further distilled into a short video—where even argumentation disappears entirely. His explanation for this process: To reach large audiences with an idea, sometimes you don’t even need theoretical justification—you just need to articulate your values clearly. These three formats serve three distinct audiences: the 15-page version for internal teams (to secure consensus); the 1-page op-ed for engaged readers (to convey positions); and the short video for the broadest audience (to transmit values).

The OpenAI Foundation Will Spend $2.5 Billion on Research, with Initial Grants Focused on Cancer Vaccines and Alzheimer’s Disease

This spring, the Foundation launched its 'AI for Alzheimer’s' initiative, awarding over $100 million in initial funding to six research institutions. Projects include collaborating with the nonprofit Arc Institute to use AI to map the disease’s 'causal atlas'; partnering with the University of Washington’s Institute for Protein Design and Mass General Brigham’s Neuroscience Institute to design novel therapeutics using AI and validate them experimentally; working with nonprofit EvE Bio and others to build open datasets predicting drug activity; collaborating with the University of California, San Francisco (UCSF) to develop new biomarkers; and testing repurposed interventions—such as orotic acid lithium and the shingles vaccine—whose patents have expired and for which pharmaceutical companies lack commercial incentive to fund clinical trials. In September, the Foundation launched its 'Public Data for Health' initiative, with over $125 million in initial grants. These include funding the University of North Carolina at Chapel Hill’s (UNC) 'Generative Immunotherapy Program' to advance personalized cancer vaccines; supporting a UCSF-led open pharmacokinetic and toxicology data project (OpenADMET) to predict how drugs cross biological membranes—including the blood-brain barrier—and organizing public prediction competitions; and establishing the Common Technical Documentation Commons (CTD Commons), a public repository collecting legacy data from bankrupt biotech firms and FDA submission documents from failed drug development programs. A third focus area targets high-burden diseases such as tuberculosis—defined by the Foundation as conditions causing over 500,000 deaths annually or equivalent health loss worldwide. Compared to the first two initiatives, publicly disclosed details about funding in this area remain scarce. It’s clear the Foundation is initially prioritizing data generation and preservation over direct investment in specific R&D pipelines. Jacob explained that breakthroughs against diseases require both intelligence *and* richer observations of the world—and in areas like Alzheimer’s, humanity still lacks sufficient knowledge to develop truly effective therapies. More diverse forms of data are urgently needed. In its official announcement, the Foundation outlined why philanthropy must step in: certain high-public-value datasets are difficult for any single institution to fund—either due to insufficient capacity or inadequate commercial incentives. Accordingly, the Foundation identifies three priority categories for funded data: 'associative data' spanning molecular, cellular, and tissue scales; 'scarce data'—once lost, irrecoverable; and 'direct data' capturing actual biological and clinical states. Using data and tools to pave the way for medical discovery is not new. In 2016, Mark Zuckerberg and Priscilla Chan’s Chan Zuckerberg Initiative (CZI) pledged $3 billion over ten years toward the goal of 'curing, preventing, or managing all diseases by the end of this century.' Its first $600 million went toward research tools and infrastructure—including founding the Biohub—to avoid betting exclusively on any single disease.

US Government, Google Join Zuckerberg-Backed Biohub in $1.8 Billion Push for AI Biology Data

16 hours ago

US government and tech giants invest $1.8 billion into Biohub to build AI datasets for biological research. The initiative aims to accelerate drug development timelines by generating predictive models from billions of cell data points within five years.

reuters

2026-10-07

US Government, Google Join Zuckerberg-Backed Biohub in $1.8 Billion Push for AI Biology Data

US Government, Google Join Zuckerberg-Backed Biohub in $1.8 Billion Push for AI Biology Data

FAQ