Median Group

Median Group (legally Median Foundation, EIN 82-5421772) is a 501(c)(3) nonprofit research organization founded in early 2018 in the San Francisco Bay Area. The organization was co-founded by Jessica Taylor, Ben Hoffman, Baeo Maltinsky, Jack Gallagher, Colleen McKenzie, and Bryce Hidysmith. Its stated mission is research on global catastrophic risks, with a focus on technological and geopolitical changes over the next century and how their most detrimental effects might be mitigated or averted.

The founding team brings backgrounds from several prominent organizations in the AI safety and effective altruism ecosystem. Jessica Taylor (President) previously worked as a mathematical researcher at the Machine Intelligence Research Institute (MIRI) and holds an MS in Computer Science from Stanford University. She led MIRI's Alignment for Advanced Machine Learning Systems program. Ben Hoffman (Secretary) worked as a researcher at GiveWell and the Open Philanthropy Project and holds an MS in Mathematics and Statistics from Georgetown University. Jack Gallagher also worked at MIRI on type theory and decision theory. Other team members include Baeo Maltinsky (BA Mathematics, UC Berkeley, former analyst at EnChroma), Colleen McKenzie (BA Computer Science and Neuroscience, Columbia, former Google product manager), Bryce Hidysmith (designer and strategist, former Numerai), and Patrick Mellor (philosophy instructor at San Francisco State University).

Median Group's most well-known work is its insight-based AI timelines model, developed by Jessica Taylor, Jack Gallagher, and Baeo Maltinsky. This interactive tool estimates when AI research might reach critical milestones by tracking the accumulation of major technical insights in AI history, calibrated against the significance of the LSTM discovery. The model assumes insights increase roughly linearly over time and allows users to adjust parameters based on their priors about the total number of insights required.

Other notable research includes a rough computational estimate of the feasibility of training AGI via deep reinforcement learning, analysis of GPU price-performance improvement rates, studies of how declining renewable energy costs may destabilize oil-dependent states, and explorations of neuroscience topics including brain computation. The organization has presented research at the Center for Human-Compatible AI at UC Berkeley.

More recent research directions have included studies of antinormative behavior in institutions, investigations of rTMS for cognitive enhancement, and development of ML tools for detecting conversational derailing patterns.

The organization received grants from the Survival and Flourishing Fund, including $98,000 in 2020 and $250,000 in 2022, both recommended by Jaan Tallinn for general operational support. Recent financial filings show declining revenue, with FY2024 reporting approximately $18,000 in revenue against $193,000 in expenses, while maintaining approximately $451,000 in net assets.