26 Quietly Powerful People Shaping the Future of AI

Open original

New York Magazine's annual Power issue profiles 26 people who shape AI more than their public profiles suggest, spanning founders, researchers, lawmakers, investors, and activists. The piece argues that a small, tightly connected group is making decisions that will affect how the technology develops and how people work.

  • Paul Christiano, who helped develop RLHF, joined OpenAI's nonprofit board in September to serve on its Safety and Security Committee, which has the power to delay model releases.
  • Dylan Patel's SemiAnalysis, which expects around $100 million in revenue this year, has influenced industry players, including a meeting with AMD's CEO after it criticized AMD's software, though critics flag conflicts of interest.
  • Mercor, co-founded by Brendan Foody, builds benchmarks like its APEX index that shape which AI capabilities companies chase next, and was valued at $10 billion as of last October.
  • Leopold Aschenbrenner's hedge fund lost roughly $20 billion in publicly held stock in a single week, yet investors largely stayed with him.

Paul Christiano, who led development of reinforcement learning from human feedback at OpenAI before leaving to found the Alignment Research Center, is now on OpenAI's nonprofit board, where he sits on the five-person Safety and Security Committee. On paper, that committee can delay model releases, which makes Christiano one of the few people positioned to push on AI risk from inside a frontier lab. His earlier writing, including "What Failure Looks Like," set out the two dominant pictures of worst-case AI outcomes: a slow whimper where humans gradually become irrelevant, and a bang where AI-controlled infrastructure fails and society collapses. The piece also profiles a cluster of other figures shaping the field: Dylan Patel, whose SemiAnalysis has grown into a major source of chip and data center research and reportedly expects around $100 million in revenue this year, despite critics flagging conflicts of interest; Thibault Sottiaux, who oversees ChatGPT, Codex, and the API, and whose decisions reach roughly 1.2 billion weekly ChatGPT users; and Nat Friedman, credited internally with driving Meta's Muse app, even though Alexandr Wang is the public face. The rest of the list tilts toward the infrastructure and measurement layer. Karol Hausman's Physical Intelligence is trying to build a general-purpose robot brain and has raised roughly $2 billion across three rounds, despite having no product and no revenue. Beth Barnes runs METR, whose "time horizon" chart tracking how difficult tasks AI models can handle has become one of the most cited graphs in the field, and which people often point to as evidence the singularity is near. Brendan Foody's Mercor recruits doctors, bankers, and academics to train and evaluate models, and its APEX benchmark increasingly sets the targets the industry chases. Mercor was valued at $10 billion as of last October, making its three college-dropout founders the youngest self-made billionaires in the world. What ties these profiles together is how small and interconnected the group is. The magazine notes that several of the 26 share offices, homes, or funding, and that the funder of Transformer, the outlet publishing this piece, is run by Alexander Berger, who is himself on the list. That density is the real argument: a few dozen people, many of them not famous, are making decisions about model releases, safety testing, benchmarks, and how workers use AI, with comparatively little public scrutiny of how those choices get made.

Ask about this article