Skip to content

Research lab

MIT Computer Science and Artificial Intelligence Laboratory

MIT research lab studying computer science and AI, spanning robotics, security, machine learning, and computational systems.

Known aliases

  • CSAIL
  • Interactive Robotics Group
  • MIT Computer Science and Artificial Intelligence Laboratory
  • MIT CSAIL

Current stories

buildOne report1 publisher

MIT CSAIL's VISTA harness lifts Claude Opus 5.0 from 40.68 to 100 on ARC-AGI-3 efficiency

MIT CSAIL's VISTA harness raised Claude Opus 5.0's ARC-AGI-3 action-efficiency score from 40.68 to 100 without retraining or replacing the model. The comparison holds the model fixed, so the lift comes from the interface built around it.

Publishers:dev.to

Reality

Evidence38
Adoption
Not scored
Hype gap+40
Incentives
Not scored
Confidence35
scienceOne report1 publisher

InstructMesh lets novices spot and repair flaws in AI-generated 3D models before printing

Novices using InstructMesh, a tool from MIT CSAIL, Google and Northeastern, found and fixed flaws in AI-generated 3D models about 90 percent of the time. An expert graded the edited designs, so how the printed objects perform is still an open question.

Publishers:news.mit.edu

Reality

Evidence40
Adoption
Not scored
Hype gap+30
Incentives70
Confidence45
scienceOne report1 publisher

AI-tuned nanoparticle recipe keeps mRNA vaccines stable for a year at room temperature

MIT researchers used an AI algorithm to reformulate mRNA vaccine nanoparticles that stay stable for a year at room temperature and two months near 38C. In mice the shots matched a Moderna-like vaccine. The lab says the change could loosen the ultracold cold chain.

Publishers:news.mit.edu phys.org

Reality

Evidence50
Adoption
Not scored
Hype gap+20
Incentives40
Confidence55
buildOne report1 publisher

Liquid AI's 279.5M drafter turns a 3.13x decode gain into 2.62x end to end

Liquid AI shipped a 279.5-million-parameter draft model for LFM2.5-VL-3B on September 24, reporting 2.30x to 3.13x faster decoding on an Apple M5 Max. Image encoding and prompt prefill are unchanged, and the company's tests did not cover quantized deployments.

Reality

Evidence55
Adoption
Not scored
Hype gap+15
Incentives70
Confidence60
scienceOne report1 publisher

A five-minute per-patient fit puts MIT's X-ray-to-CT registration under a millimeter

Existing 2D/3D registration models align well for some patients and fail for others. MIT's xvr, out today in Nature, fits itself to one patient in about five minutes, then matches that patient's X-rays to their 3D scan in seconds.

Publishers:news.mit.edu

Reality

Evidence58
Adoption
Not scored
Hype gap+30
Incentives70
Confidence55
productOne report1 publisher

MIT's TONTOU attack reaches protected Linux memory through a two-instruction window

CSAIL researchers timed an interrupt to land inside the gap AMD's saferet defense leaves open, then pulled the root password file off a current Linux kernel. AMD has shipped a mitigation. Intel's picture is messier.

Reality

Evidence38
Adoption30
Hype gap+26
Incentives55
Confidence44