Arb Research is a small research consultancy producing rigorous, independent analysis on AI safety, forecasting, and related topics for funders and organizations in the effective altruism ecosystem.
Arb Research is a small research consultancy producing rigorous, independent analysis on AI safety, forecasting, and related topics for funders and organizations in the effective altruism ecosystem.
People
Updated 05/18/26Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 05/18/26Arb Research is a small, remote-first research consultancy founded around 2021 by Misha Yagudin and Gavin Leech. The firm operates at the intersection of AI safety, forecasting, and policy, providing high-quality independent research to clients in the effective altruism and AI safety ecosystems. Current leadership includes Gavin Leech (co-founder, AI PhD from University of Bristol) and Charles Dillon (former senior portfolio manager at Millennium). The team of roughly nine people spans machine learning PhDs, superforecasters, and policy researchers. Arb's most notable recurring project is the Shallow Review of AI Safety, a comprehensive annual survey mapping active research agendas across the AI safety field. The 2024 edition reviewed over 800 links and consulted dozens of researchers. The firm also produced a 90-page survey of AI's likely effects across ten perspectives including ML, governance, economics, and ethics, and contributed hundreds of AI forecasting questions for Metaculus's Future of AI series. In 2025, Gavin Leech co-authored The Scaling Era: An Oral History of AI, 2019-2025 with Dwarkesh Patel, published by Stripe Press. Other notable work includes impact assessments of AI Safety Camp, research on questionable research practices in machine learning, studies on AI bias against human-generated content (published in PNAS), and developing Talleyrand, an AI platform for scenario analysis. Arb has received multiple grants from Open Philanthropy, including $25,000 for staff support, $70,000 to support an AI Governance Retreat for professionals in the field, and $5,000 for an updated AI safety shallow review. The firm operates as a consultancy with fee-for-service clients rather than primarily as a donation-funded nonprofit.
Theory of Change
Updated 05/18/26Arb believes that improving the quality and accessibility of research on AI safety and AI forecasting helps funders, policymakers, and researchers make better decisions. By producing rigorous, independent landscape reviews and evaluations, Arb reduces information asymmetries in the AI safety field, enabling grants and organizational resources to flow to higher-impact work. Their forecasting work improves collective epistemic accuracy about AI trajectories, which feeds into better strategic planning by safety organizations. By identifying questionable practices in ML evaluation, they help establish higher standards for AI research quality, indirectly improving the reliability of safety-relevant benchmarks and assessments.
Grants Received
Updated 05/18/26Projects
Updated 05/18/26A December 2023 Arb Research report commissioned by Open Philanthropy that compiles 732 forecasting questions about future AI capabilities, safety, governance, and timelines, intended to feed into forecasting tournaments and research pipelines such as Metaculus’s Future of AI series.
A January 2024 Arb Research impact evaluation of AI Safety Camp that uses surveys and participant tracking to estimate what fraction of attendees become AI safety researchers, the cost per additional researcher produced, and other outcomes such as publications and placements.
An Arb Research project, funded by Emergent Ventures, that uses AI‑assisted methods and manual review to catalog over 100 cases where UK and US regulators disagree on whether to approve specific medicines, highlighting systematic cross‑country differences in drug availability.
Discussion
No comments yet. Be the first to share your thoughts.