Divya Shanmugam
@dmshanmugam
building reliable models from unreliable data — currently postdoc @cornell_tech, formerly phd @MIT_CSAIL
Today we're releasing Community Alignment - the largest open-source dataset of human preferences for LLMs, containing ~200k comparisons from >3000 annotators in 5 countries / languages! There was a lot of research that went into this... 🧵
Are LLMs correlated when they make mistakes? In our new ICML paper, we answer this question using responses of >350 LLMs. We find substantial correlation. On one dataset, LLMs agree on the wrong answer ~2x more than they would at random. 🧵(1/7)
When you walk into the ER, you could get a doc: 1. Fresh from a week of not working 2. Tired from working too many shifts I’ve been both and believe me, they're different! And we can tell the difference, just from notes they write. Paper @NatureComms nature.com/articles/s4146…
I can’t believe I’m saying this: our work received a Best Paper Award at #CHIL2025!! So so excited and grateful 🥰 Looking forward to day 2 of the conference with these awesome people :)
I’m really excited to share the first paper of my PhD, “Learning Disease Progression Models That Capture Health Disparities” (accepted at #CHIL2025)! Link and summary in thread✨1/
I wrote about science cuts and my family's immigration story as part of the McClintock Letters initiative. Haven't yet placed it in a Houston-based newspaper but hopefully it's useful here gargnikhil.com/posts/20250614…
For folks at @FAccTConference, we're very excited to present this work tomorrow at the 10:45am "Bias Across Languages" session!
🎉 Excited to present our paper at @FAccTConference , co-authored with @JieboLuo @jiank_uiuc @allisonkoe : “Characterizing Bias: Benchmarking Large Language Models in Simplified versus Traditional Chinese” revealing 🤖LLM performance disparities. 📄 Link: arxiv.org/abs/2505.22645
Had (what I thought) was an interesting human-AI interaction this morning — was using iNaturalist to ID a bug I had seen outside my apartment, and none of the top-K suggestions seemed to match …
new decade, same verse
It’s a hefty 206-page research paper, and the findings are concerning. "LLM users consistently underperformed at neural, linguistic, and behavioral levels" This study finds LLM dependence weakens the writer’s own neural and linguistic fingerprints. 🤔🤔 Relying only on EEG,…
I had a great conversation with the @siegelendowment about how various technologies—from complex AI tools to basic self checkout—are changing the nature of service jobs for millions of Americans🛒 Give it a read to learn more about my recent work! :) siegelendowment.org/insights/the-h…
I've recently put together a "Fairness FAQ": tinyurl.com/fairness-faq. If you work in non-fairness ML and you've heard about fairness, perhaps you've wondered things like what the best definitions of fairness are, and whether we can train algorithms that optimize for it.
What happens in SAIL 2025 stays in SAIL 2025 -- except for these anonymized hot takes! 🔥 Jotted down 17 de-identified quotes on AI and medicine from medical executives, journal editors, and academics in off-the-record discussions in Puerto Rico
Proud of my three daughters — all doctors, but very different fields! 🎉🥰🎉Congratulations especially to Nithya, a new graduate of the Emory School of Medicine.