🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Great news! If you haven't had the chance to submit your application for MLESI24, the deadline has been extended to April 12th. Make sure to get your applications in! ➡️ buff.ly/49hRlAd

Sendhil Mullainathan Massachusetts Institute of Technology (MIT) coming Nov. 11 - 13 to Cornell University as esteemed Messenger Lecturer, Center for Data Science for Enterprise and Society Distinguished Lecturer, co-sponsored w/ Law, Econ, and Policy seminar, CS Colloquium, ORIE Colloquium events.cornell.edu/event/messenge…

In our new PNASNews paper, across 21 experiments with 23,000+ participants, we identify a critical distortion that shapes decisions involving tradeoffs: we find that people systematically overweight quantified information in such decisions. Paper: pnas.org/doi/10.1073/pn… 🧵

Economic valuations fluctuate in ways empirical research cannot fully explain What information are we missing? Economic theories emphasize the role of hard-to-quantify beliefs and perceptions My job market paper develops algorithms + measurement to quantify perceptions of firms

Also see: x.com/StatMLPapers/s… Jiaxun Li, Vinod Raman, and Ambuj Tewari generalize Jon Kleinberg and 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧’s results to non-countable collections and non-uniform generatability!

A follow-up to interesting work below: arxiv.org/abs/2411.09642 TLDR: ➤ Two essential properties of a good language model are no hallucination and no mode-collapse ➤ Jon Kleinberg-𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 asked: Can a model achieve both? ➤ We show it is impossible: x.com/AnayMehrotra/s…

Has anyone been running more world model metrics on o1/o3 since release (does ARC count?)? This paper popped into my head again recently and since they have a really nice open source project, I ran their logic puzzle metric on o1-mini.

#QJE Feb 2025, #12, “Do Financial Concerns Make Workers Less Productive,” by Kaur (Supreet Kaur), Mullainathan (𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧), Oh (Suanna Oh), and Schilbach (@FrankSchilbach): doi.org/10.1093/qje/qj…

On the use of large language models to solve prediction and estimation problems - amazing paper by 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 and colleagues. tylercowen nber.org/papers/w33344

Our article on using LLMs to improve health equity is out in NEJM AI! 85% of equity-related LLM papers focus on *harms*. But equally vital are the equity-related *opportunities* LLMs create: detecting bias, extracting structured data, and improving access to health info.

More than one year ago, 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧, interviewed at #Capitalisnt, forecasted that there were no barriers to entry in AI. The success of DeepSeek proved him right. Listen to #Capitalisnt, where we discuss today about next year’s news capitalisnt.com/episodes/who-c…

We have fun and talk about DeepSeek in the second episode of a new podcast on the frontier of AI and economic productivity organized by the great Ajay Agrawal. Check it out. insights.intrepidgp.com//p/dm-ep2-deep…

Next session of the Economics of AI series 📅 19 March 17:00 CET How is AI transforming the scientific method? Join 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 Massachusetts Institute of Technology (MIT) as he explores Science in the Age of Algorithms, with discussion by Annie Liang (Northwestern & CEPR). 🔗 Register now: cepr.org/events/cepr-we…

MIT professor Sendhil Mullainathan calls Artificial General Intelligence (AGI) — a form of AI that would possess human-level cognitive abilities — "a shockingly unambitious goal."

Next session of the Economics of #AI series 📅 19 March 17:00 CET How is AI transforming the scientific method? Join 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 Massachusetts Institute of Technology (MIT) as he explores Science in the Age of #Algorithms, with discussion by Annie Liang (Northwestern & CEPR). 🔗 Register now: cepr.org/events/cepr-we…

In the The Wall Street Journal, our research affiliate 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 argues that instead of setting benchmarks to encourage automation of human tasks, we should develop AI as a "bicycle for the mind" — a technology that can help us do things better. wsj.com/tech/ai/ai-job…

Can you guess which of these two questions an LLM needs more thinking tokens to get right? 🧵

CEPR Webinar Series on the Economics of Artificial Intelligence 3: Science in the Age of Algorithms by 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 with Anton Korinek & Anni Liang youtu.be/p4mF4c1ZLv

The WORLD'S FIRST legally authorized AI medical device approved to make CLINICAL DECISIONS on skin cancer WITHOUT HUMAN REVIEW. This is the tip of the iceberg. So much coming to improve accessibility and speed of healthcare. Listen on Derby Mill. 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 Richard Sutton

AI models appear to mimic the real world. But how can we tell if they truly understand it? Excited to announce the ICML 2025 Workshop on Assessing World Models! Working on related questions? Submit a paper (max. 4 pages) by May 20!