MIT LIDS

Planning complex visual tasks with #genAI New system from Yilun Hao, Yongchao Chen, Yang Zhang, Chuchu Fan plans long-term visual tasks 2x as well as existing methods. This could help robots navigate changing environments or coordinate multirobot assembly. bit.ly/4t2vczN

🧬 AI for a fuller picture of the cell A new #AI framework identifies which cellular signals are unique to a measurement method — and which are shared across modalities. The approach helps researchers better understand cell states and disease mechanisms. bit.ly/4ufvuV0

Have you noticed how navigation apps include walking & waiting for public transit, but excludes parking & walking for driving? After being late a few times 😅, we finally did. We got curious: what if these apps account for parking?

Reducing emissions — and parking frustration 🚗♻️ MIT LIDS researchers developed a parking-aware navigation system that guides drivers to lots balancing proximity + availability. The result? Up to 35 minutes saved and more accurate travel time estimates. bit.ly/3P5QYne

Can personalization make LLMs too agreeable? Research from @MITLIDS Shomik Jain, Charlotte Park, and PI Ashia Wilson finds that long-term conversations can cause LLMs to mirror users’ views — potentially reducing accuracy & reinforcing echo chambers. More: bit.ly/3OqbC1b

We’ve recently made the repos public! More info + paper coming soon from @MITLIDS @GioeleZardini lab ~~ github.com/mit-zardini-la… github.com/mit-zardini-la…

MIT research finds platforms ranking the latest #LLMs can be surprisingly fragile — removing just a handful of crowdsourced votes can flip which model tops the leaderboard. The work underscores a need for more rigorous strategies to evaluate model rankings bit.ly/4csuBC8

Where the ocean and atmosphere communicate 🌊🌬️ In a recent profile, LIDS PI @abigailbodner explores how ocean turbulence shapes climate patterns—and how math, Earth science, and AI come together in her research. Read the profile in MIT Spectrum: bit.ly/4qWJv8d

Moving beyond overly aggregated #ML metrics. New research from LIDS postdoc Olawale Salaudeen, PI Marzyeh Ghassemi, and collaborators reveals how commonly used, highly aggregated ML metrics can hide serious failures — and offers a method to correct them. bit.ly/4k5LMLE

Health data breaches are becoming routine: 747 major incidents recorded by HHS in the past 24 months. New MIT LIDS research shows that foundation #AI models trained on de-identified EHRs can still memorize and expose sensitive patient data. bit.ly/4sFzZYx

AI will transform every sector of human society in fundamental ways. Is there a case to be made for how it can be a force for good? Tell us what you think the future holds. Over $20K in prizes will be awarded. Open to MIT students only. Submit by 2/8/2026. computing.mit.edu/FutureComputin…

A new paper in @LancetRH_Americ provides the most comprehensive study of physical restraint use in U.S. ICUs to date. The study analyzes restraint use across 51k+ patients over 14 years. thelancet.com/journals/lanam…

#AI is often criticized for its growing energy use—but it can also help modernize the power grid. In a 3Q with MIT News, LIDS PI Priya Donti explains how AI can boost grid efficiency, improve resilience to extreme weather, and enable more renewable energy. bit.ly/4pGrwkM

📢 Asu Ozdaglar and I are hiring a postdoc for next year to work on adaptive, ethically-informed decision systems, human–AI collaboration, and decision-making under uncertainty. More info here: charapodimata.com/postdoc_announ….

New prognostic tool could help clinicians identify high-risk lymphoma patients earlier—and tailor treatment to improve survival. MIT LIDS researchers show how early relapse strongly predicts outcomes using Synthetic Survival Controls. Published in Blood. bit.ly/45zomYY

DAAAM!! "Describe Anything Anywhere at Any Moment". State of the art approach to provide spatio-temporal memory to robots and agents. Powered by VLMs and scene graphs. Directly suitable for LLM queries. great work by Nicolas Gorlo and Lukas Schmid! nicolasgorlo.com/DAAAM_25/

A new op-ed from Richard Locke, Dean of @MITSloan, highlights research from our co-directors @DAcemogluMIT, @baselinescene, and @davidautor and our affiliate @m_sendhil on choosing a pro-human path for AI. mitsloan.mit.edu/ideas-made-to-…

Researchers in EECS and AeroAstro developed a tiny aerial robot that can fly with speed and agility comparable to some insects. The research opens the door to future bug-sized robots that could aid in search-and-rescue missions. news.mit.edu/2025/mit-engin…

A smarter way for LLMs to reason 🧠 MIT LIDS researchers developed a method that lets LLMs dynamically adjust how much computation they use based on a question’s difficulty—boosting efficiency without sacrificing performance. bit.ly/4p00Knf

AI is too important to leave to computer scientists. MIT’s Sendhil Mullainathan is using algorithms to transform economics and decision-making, Bob Simison writes in F&D magazine. imf.org/en/publication…