Improving AI models’ ability to explain their predictions
A new approach could help users know whether to trust a model’s predictions in safety-critical applications like health care and autonomous driving.
A “ChatGPT for spreadsheets” helps solve difficult engineering challenges faster
The approach could help engineers tackle extremely complex design problems, from power grid optimization to vehicle design.
Engineering confidence to navigate uncertainty
In 16.85 (Design and Testing of Autonomous Vehicles), AeroAstro students build software that allows autonomous flight vehicles to navigate unknown environments.
New method could increase LLM training efficiency
By leveraging idle computing time, researchers can double the speed of model training while preserving accuracy.
Mixing generative AI with physics to create personal items that work in the real world
To help generative AI models create durable, real-world accessories and decor, the PhysiOpt system runs physics simulations and makes subtle tweaks to its 3D blueprints.
AI to help researchers see the bigger picture in cell biology
By providing holistic information on a cell, an AI-driven method could help scientists better understand disease mechanisms and plan experiments.
Chip-processing method could assist cryptography schemes to keep data secure
By enabling two chips to authenticate each other using a shared fingerprint, this technique can improve privacy and energy efficiency.
Exposing biases, moods, personalities, and abstract concepts hidden in large language models
A new method developed at MIT could root out vulnerabilities and improve LLM safety and performance.
A neural blueprint for human-like intelligence in soft robots
An AI control system co-developed by SMART researchers enables soft robotic arms to learn a broad set of motions once and adapt instantly to changing conditions without retraining.
Parking-aware navigation system could prevent frustration and emissions
By minimizing the need to drive around looking for a parking spot, this technique can save drivers up to 35 minutes — and give them a realistic estimate of total travel time.
Personalization features can make LLMs more agreeable
The context of long-term conversations can cause an LLM to begin mirroring the user’s viewpoints, possibly reducing accuracy or creating a virtual echo-chamber.
Study: Platforms that rank the latest LLMs can be unreliable
Removing just a tiny fraction of the crowdsourced data that informs online ranking platforms can significantly change the results.
Helping AI agents search to get the best results out of large language models
EnCompass executes AI agent programs by backtracking and making multiple attempts, finding the best set of outputs generated by an LLM. It could help coders work with AI agents more efficiently.
Brian Hedden named co-associate dean of Social and Ethical Responsibilities of Computing
He joins Nikos Trichakis in guiding the cross-cutting initiative of the MIT Schwarzman College of Computing.
Antonio Torralba, three MIT alumni named 2025 ACM fellows
Torralba’s research focuses on computer vision, machine learning, and human visual perception.