Mark Zuckerberg

CEO & Founder — Meta

Profile

Mark Zuckerberg is the founder and CEO of Meta, and — depending on who you ask — either the most important open-source advocate in AI or a latecomer who had no choice but to open-weight his way out of a strategic hole. Either way, the result is the same: Llama reset expectations for what you could run, fine-tune, and build on without begging a closed-model provider for API credits.

After the metaverse bet burned tens of billions and Reality Labs kept bleeding, Zuckerberg executed one of the sharpest corporate pivots in recent memory. Meta AI went from a sprawling research shop — FAIR, run by Yann LeCun, long skeptical of LLMs as the path to intelligence — to the engine of a company-wide strategy. Llama 1 leaked in early 2023, Llama 2 shipped with a commercial license, and by Llama 3 and 3.1 (the 405B model) Meta was releasing frontier-ish weights for free. That single decision shaped the entire open-weights ecosystem: Mistral, Qwen, DeepSeek, and countless fine-tunes exist in a world Zuck made viable.

In 2025 he went harder. Meta Superintelligence Labs was stood up with a nine-figure hiring spree, Alexandr Wang was brought in as Chief AI Officer via a $14B stake in Scale AI, and Zuckerberg personally poached researchers from OpenAI, Anthropic, and Google DeepMind with comp packages that reset industry norms. Capex guidance for AI infrastructure crossed $60B. Llama 4 shipped with mixed reviews — the open-source honeymoon cooled as licensing got more restrictive and benchmarks looked gamed — but Meta AI is now embedded in WhatsApp, Instagram, and Facebook, putting an assistant in front of roughly three billion people.

For developers, Zuckerberg matters because he’s the reason you can download weights at all. Whether that stays true — or whether Meta closes the gates once the strategic value of “open” runs out — is the question hanging over the next couple of model releases.

Key Articles & Papers

Open Source AI Is the Path Forward 2024 — The manifesto accompanying Llama 3.1's release. Zuckerberg argues open-source AI is good for developers, Meta, and the world — and lays out why closed models shouldn't be the default. LLaMA: Open and Efficient Foundation Language Models 2023 — The original Llama paper. Smaller, efficient models trained on public data — the release that accidentally kicked off the open-weights era after the weights leaked. Llama 2: Open Foundation and Fine-Tuned Chat Models 2023 — First Llama release with a commercial license. The paper that made building products on open weights legally viable for most companies. The Llama 3 Herd of Models 2024 — Technical report for Llama 3, including the 405B model — the first open-weights model in genuine frontier territory. Founder's Letter: Building the Next Computing Platform 2021 — The Meta rebrand announcement. Worth reading to understand how much the company has turned since — AI now eats the bets this letter was about.

Videos

Controversies

Cambridge Analytica (2018): Facebook allowed a third-party app to harvest data from ~87 million users, which was used for political profiling. Led to Zuckerberg’s Congressional testimony and ongoing regulatory scrutiny of Meta’s data practices.
Teen mental health & Section 230: Internal research leaked by whistleblower Frances Haugen showed Instagram’s harms to teen users were known internally. Meta has been sued by dozens of U.S. state attorneys general. Relevant to AI because the same engagement-optimization instincts now shape how Meta AI is deployed.
“Open source” labeling: The OSI and researchers have argued Llama’s license isn’t actually open source — it restricts commercial use above certain thresholds and doesn’t release training data. Critics call it “open-washing”; Meta has pushed back but hasn’t changed the license.
Talent poaching in 2025: The reported $100M+ offers to OpenAI and Anthropic researchers drew public pushback from Sam Altman and others, raising questions about whether buying a superintelligence team works.