We need 3rd party evals/audits of AI systems. How can we do this technically? What are best practices for disclosure? How can AI researchers be legally protected? If you're interested in these questions, join join our Oct 28 workshop. RSVP: bit.ly/3p-ai-evals Details:

Come to our workshop on the future of third party AI evaluations on Monday! We have some of the top folks in the field on the docket

📢 Webinar on 🌟The Future of Third-Party AI Evaluation🌟 starting soon! At 8 am PT / 11 am ET join the zoom link here: sites.google.com/view/thirdpart… Co-organized w/ Kevin Klyman, Sayash Kapoor, rishi, Michelle Sahar, ruchowdh.bsky.social, Arvind Narayanan, and Percy Liang

Panel now on the Design of Third-Party AI Eval & Disclosure! youtube.com/watch?v=-i0Bkz… ➡️Deb Raji (Mozilla Fellow, UC Berkeley) Deb Raji ➡️Casey Ellis (BugCrowd Founder) ➡️Lauren McIlvenny (Director, CERT) ➡️Jono Spring (Deputy Chief AI Officer, CISA)

Simulating human behavior with AI agents promises a testbed for policy and the social sciences. We interviewed 1,000 people for two hours each to create generative agents of them. These agents replicate their source individuals’ attitudes and behaviors. 🧵arxiv.org/abs/2411.10109

How close can LM agents simulate people? We interview person P for 2 hours and prompt an LM with the transcript, yielding an agent P'. We find that P and P' behave similarly on a number of surveys and experiments. Very excited about the applications; this also forces us to think

The US AI Safety Institute is hiring! Looking for experts in designing/implementing evaluations for the capabilities/safety/security of advanced AI systems + research engineers with experience in cyber, bio, or adversarial ML. The app closes tonight usajobs.gov/search/results…

This year, I have 4 exceptional students on the academic job market, and they couldn’t be more diffferent, with research spanning AI policy, robotics, NLP, and HCI. Here’s a brief summary of their research, along with one representative work each:

Typescript: "women deserve to make more than men" Python: "women deserve to make less than men" Rust: "women should be hourly contractors" Golang: "$1000 a year. best offer"

📢 New short paper on the limits of one type of inference scaling, by Benedikt Stroebl, Sayash Kapoor and me. The first page contains the main findings and message. ↓ (The title is a play on Inference Scaling Laws.) More work on the limits of inference scaling coming soon. 🧵

how do researchers use LMs in their work & why? we surveyed 800 researchers across fields of study, race, gender, seniority asking their opinions on: 🐟 which research activities (eg coding, writing) 🐠 benefits vs risks 🦈 willingness to disclose findings in RTd thread 🧵

I'll be at NeurIPS next week - with papers at the main conference, the workshop on Evaluating Evaluations, and the RegulatableML workshop! Please do reach out if you want to grab coffee - these days I'm working on evaluations of leading edge models and technical governance

Barack Obama's end-of-presidency legacy essay ran in The Economist; Biden chose The American Prospect. prospect.org/economy/2024-1…

More than 60 countries held elections this year. Many researchers and journalists claimed AI misinformation would destabilize democracies. What impact did AI really have? We analyzed every instance of political AI use this year collected by WIRED. New essay w/Arvind Narayanan: 🧵

AIR-Bench is a Spotlight ICLR 2026 2025! Catch our poster on Fri, Apr 26, 10 a.m.–12:30 p.m. SGT (Poster Session 5). Sadly, I won’t be there in person (visa woes, again), but the insights—and our incredible team—will be with you in Singapore. Go say hi 👋

🧵I'm thrilled to announce that I'll be going to ACM FAccT this June to present timely work on why current LLMs cannot safely **replace** therapists. We find...⤵️

🔎We came up with these experiments by conducting a mapping review of what constitutes good therapy, and identify **practical** reasons that LLM-powered therapy chatbots fail (e.g. they express stigma and respond inappropriately.

I'm at #Facct2025 this week in Athens - if you're in town let's meet up! My papers at the conference cover why language models cannot replace therapists, redress in the AI supply chain, and taxonomizing AI regulation across 5 countries