A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
MIT researchers developed FloatForm, a swarm of small aquatic robots that snap together like ants forming a raft, assembling into reconfigurable structures on the water.
New dataset of 30,000-plus competition math problems from 47 countries gives AI researchers a harder test — and students worldwide a better training ground.
A new training method improves the reliability of AI confidence estimates without sacrificing performance, addressing a root cause of hallucination in reasoning models.
Researchers use control theory to shed unnecessary complexity from AI models during training, cutting compute costs without sacrificing performance.
CSAIL researchers find even “untrainable” neural nets can learn effectively when guided by another network’s built-in biases using their guidance method.
MIT CSAIL and LIDS researchers developed a mathematically grounded system that lets soft robots deform, adapt, and interact with people and objects, without violating safety limits.
The coding framework uses modular concepts and simple synchronization rules to make software clearer, safer, and easier for LLMs to generate.
MIT CSAIL and McMaster researchers used a generative AI model to reveal how a narrow-spectrum antibiotic attacks disease-causing bacteria, speeding up a process that normally takes years.
