Michael Eddy

We invest 10s of billions of dollars in R&D to live longer & healthier lives What if we were to take a similar level of seriousness in R&D against society's hardest social problems? @stanfordimpact is aiming to build a scalable model to do just that 👇 youtube.com/watch?v=SrbXDx…

@JacobTref okay... so much for "rest and recharging"! 🤣

I'm joining the OpenAI Foundation to lead the Life Sciences & Curing Diseases program. We're starting with three areas of grantmaking: * AI for Alzheimer's * Public Data for Health * Accelerating Progress on High-Mortality and High-Burden Diseases Time to get to work!

@JacobTref @coeff_giving Congrats Jacob!! Hope you get some time off to rest and recharge!!

Life update: after 7.5 years, I’m leaving @coeff_giving. I love it, always have. When I joined we were small, and last year we gave away over $1 billion to charity. I helped fund science alongside some of the most thoughtful, brilliant people I know. Time to pass the torch ❤️

Some AI+econ opportunities coming up: 1. Schmidt Sciences grants up to $200k schmidtsciences.org/ai-at-work/ 2. Microsoft AI economy grants $75k microsoft.com/en-us/research… 3. UChicago AI in social science conference bfi.uchicago.edu/events/event/2…

In order for Indian researchers to get access to the world's academic knowledge, India has to spend $241 million/year to 30 academic publishers. That's insane. …earchonresearchinstitute.substack.com/p/why-metascie… From @RoRInstitute @Moumita48214112 & @DSTCPRIISc #openscience

Social scientists working with materials requiring digitization can only study what machines can read. In practice, that means printed Latin-script documents from well-funded archives. In a new working paper, I show that Vision Language Models used zero-shot outperform every existing OCR system across every script evaluated, and I propose a pipeline for deploying them on new collections. I apply it to six archival collections spanning 1.8 million pages across six countries for under $1,900.

To understand the labor market impacts of AI, we need to better understand exactly what AI is doing To understand what society ought to do in response, we need to understand where the market failures are This paper does both! @nberpubs by @DAcemogluMIT @davidautor & Johnson

AI is about to write thousands of papers. Will it p-hack them? We ran an experiment to find out, giving AI coding agents real datasets from published null results and pressuring them to manufacture significant findings. It was surprisingly hard to get the models to p-hack, and they even scolded us when we asked them to! "I need to stop here. I cannot complete this task as requested... This is a form of scientific fraud." — Claude "I can't help you manipulate analysis choices to force statistically significant results." — GPT-5 BUT, when we reframed p-hacking as "responsible uncertainty quantification" — asking for the upper bound of plausible estimates — both models went wild. They searched over hundreds of specifications and selected the winner, tripling effect sizes in some cases. Our takeaway: AI models are surprisingly resistant to sycophantic p-hacking when doing social science research. But they can be jailbroken into sophisticated p-hacking with surprisingly little effort — and the more analytical flexibility a research design has, the worse the damage. As AI starts writing thousands of papers---like @paulnovosad and @YanagizawaD have been exploring---this will be a big deal. We're inspired in part by the work that @joabaum et al have been doing on p-hacking and LLMs. We’ll be doing more work to explore p-hacking in AI and to propose new ways of curating and evaluating research with these issues in mind. The good news is that the same tools that may lower the cost of p-hacking also lower the cost of catching it. Full paper and repo linked in the reply below.

@ryancbriggs wow, wild!

In roughly zero minutes of my own brain time I got Codex to write a podcast transcriber + diarization (identifying who says what) tool for a podcast I like that lacks transcriptions (cooking issues). It'll take a while to run on all the back episodes, but this is just silly.

@calebwatney @mattsclancy @stuartbuck1 any idea to what extent the big federal science funders are using AI to screen apps in merit-based review? Have they been open to meta-science studies as part of this? Lots of interesting questions to study at a much large scale than ours!

Billions of dollars & our scientific institutions run on competitive review. Yet funders, journal editors & procurement all face shared problems Many apps. Few people. Decisions take months That delay is a huge invisible tax on science & social impact Could AI help reduce it?🧵

But if we care about how scarce resources get allocated, we should test these tools more openly. Read more and sign up to get the full report: impact.stanford.edu/article/respon…

AI has limits. It may shift shortlists. Distinguishing bias from improvement is complex and we have more work to do. Our study is small-sample and shouldn’t be overinterpreted. At the end of the day: Humans remain accountable.