📢 My wonderful colleagues Nina Roussille and Jacob Moscona (and I) are jointly hiring predoctoral research fellows to work in the exciting research environment MIT Economics! Please spread the word! 🙏 Econ RA Listings

Calling all postgrads 🔊 Apply for the Machine Learning in Economics Summer Institute 2024 to work with leading researchers from top institutes to push forward cutting-edge research ideas. Find more information and the application at the link below: buff.ly/3mm92Ls

With the Spring semester starting at Massachusetts Institute of Technology (MIT), I'm excited to be teaching two new courses on machine learning and economics with 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 (one for PhD students and one for undergraduates). We'll be posting lecture slides here for anyone to follow along: economics.mit.edu/people/faculty…

Applications for the Machine Learning in Economics Summer Institute (MLESI) are open! Grad students, apply and come hang out in Chicago for a week to learn about ML and Econ.

Calculating the social return on algorithmic interventions, specifically their Marginal Value of Public Funds, across multiple domains — regulation, criminal justice, medicine, and education, from Jens Ludwig, 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧, and Ashesh Rambachan nber.org/papers/w32125

ICYMI: Building foundation models to model career transitions

Reminder: there's an awesome new interdisciplinary conference on Economics and AI. The paper submission deadline is this Sunday, Feb 25!

Are algorithms overhyped? New research from Harris Policy's Jens Ludwig, Chicago Booth's 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧, & Ashesh Rambachan on the effectiveness of algorithms for addressing public policy problems suggests the opposite: ow.ly/HhFZ50QGI9o

Thank you Schmidt Sciences for the support! Such an honor to be in this incredible cohort of researchers. Excited to work towards ensuring AI functions as intended in real-world settings. Berkeley Statistics Berkeley Computing, Data Science, and Society schmidtsciences.org/ai2050-early-c…

Don't forget! Applications for the MLESI 2024 cohort close in one week. Don't miss this opportunity to grow under the guidance of industry experts from various top institutions. Register at the link below. ⬇️

For all of our machine learning and economics researchers—don't forget to submit your papers for the first Machine Learning in Economics Summer Conference (MLESC) by this Friday, 03/29. We would love to read your work! buff.ly/3IVaY5T

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

This is the last weekend to work on your MLESI applications! 7 days left to apply. ⏰ buff.ly/49hRlAd

Recently accepted by #QJE, “Identifying Prediction Mistakes in Observational Data,” by Ashesh Rambachan (Ashesh Rambachan): doi.org/10.1093/qje/qj…

Increasing numbers of people claim that current AI systems "understand" and "reason" through the problems they are given. Or at least are on the cusp of it, so that Artificial General Intelligence is right around the bend. I don't think that's the case. Here is why.

This was an extremely fun paper to be a part of! How can we evaluate whether a transformer has uncovered the right "world model"?

#QJE May 2024, #6, “Identifying Prediction Mistakes in Observational Data,” by Ashesh Rambachan (Ashesh Rambachan): doi.org/10.1093/qje/qj…

Humans decide which tasks to use general-purpose large language models for, “so we have to take the human in the loop into account,” says Ashesh Rambachan. A new method evaluates a model based on its alignment with a human’s beliefs about its capabilities. mitsha.re/ROUX50SIki0

How can the same LLM pass AP math and also claim 9.10 > 9.9? Our #ICML2024 paper: LLMs are wrong in ways that people can't predict. Benchmarks can be misleading because people decide how LLMs are used. Solution: measure LLM alignment with the "human generalization function" 🧵

LLMs don’t behave like people, even though we may expect them to. A new study from LIDS PIs Ashesh Rambachan and 𝐒𝐞𝐧𝐝𝐡𝐢𝐥 𝐌𝐮𝐥𝐥𝐚𝐢𝐧𝐚𝐭𝐡𝐚𝐧 shows someone’s beliefs about an LLM play a significant role in the model’s performance and are important for how it is deployed. bit.ly/3y2zmRm