Teaching AI models what they don’t know
A team of MIT researchers founded Themis AI to quantify AI model uncertainty and address knowledge gaps.
Computer Science and Artificial Intelligence Laboratory (CSAIL)
Teaching AI models the broad strokes to sketch more like humans do
SketchAgent, a drawing system developed by MIT CSAIL researchers, sketches up concepts stroke-by-stroke, teaching language models to visually express concepts on their own and collaborate with humans.
Eight with MIT ties win 2025 Hertz Foundation Fellowships
The fellowships recognize doctoral students who have “the extraordinary creativity and principled leadership necessary to tackle problems others can’t solve.”
Imaging technique removes the effect of water in underwater scenes
The color-correcting tool, known as “SeaSplat,” reveals more realistic colors of underwater features.
Study shows vision-language models can’t handle queries with negation words
Words like “no” and “not” can cause this popular class of AI models to fail unexpectedly in high-stakes settings, such as medical diagnosis.
System lets robots identify an object’s properties through handling
With a novel simulation method, robots can guess the weight, softness, and other physical properties of an object just by picking it up.
Hybrid AI model crafts smooth, high-quality videos in seconds
The CausVid generative AI tool uses a diffusion model to teach an autoregressive (frame-by-frame) system to rapidly produce stable, high-resolution videos.
Novel AI model inspired by neural dynamics from the brain
New type of “state-space model” leverages principles of harmonic oscillators.
Making AI models more trustworthy for high-stakes settings
A new method helps convey uncertainty more precisely, which could give researchers and medical clinicians better information to make decisions.
The MIT-Portugal Program enters Phase 4
New phase will support continued exploration of ideas and solutions in fields ranging from AI to nanotech to climate — with emphasis on educational exchanges and entrepreneurship.
“Periodic table of machine learning” could fuel AI discovery
Researchers have created a unifying framework that can help scientists combine existing ideas to improve AI models or create new ones.
3D modeling you can feel
TactStyle, a system developed by CSAIL researchers, uses image prompts to replicate both the visual appearance and tactile properties of 3D models.
Making AI-generated code more accurate in any language
A new technique automatically guides an LLM toward outputs that adhere to the rules of whatever programming language or other format is being used.
Training LLMs to self-detoxify their language
A new method from the MIT-IBM Watson AI Lab helps large language models to steer their own responses toward safer, more ethical, value-aligned outputs.
New method efficiently safeguards sensitive AI training data
The approach maintains an AI model’s accuracy while ensuring attackers can’t extract secret information.