What Does Chain-of-Thought Entropy Measure? A Channel Audit of Scaffolding, Routing, and Content
Abstract page for arXiv paper 2609.25039: What Does Chain-of-Thought Entropy Measure? A Channel Audit of Scaffolding, Routing, and Content
News index / Research
AI research papers and lab updates
Follow new papers, research lab posts, benchmarks, academic ML work, and deep learning breakthroughs from leading research groups and journals.
Reading queue
05 entries"As a Language Model...": Chat Template Switches LLM Self-Referential Voice and Activation Steering Reproduces It
cs.LG updates on arXiv.org · Research · 5h ago
Poitras Center to fuel early careers of 50 young scientists dedicated to psychiatric disorders research
MIT News - Artificial intelligence · Research · 14h ago
Watch scientists decipher burnt scrolls without unrolling them
Machine learning : nature.com subject feeds · Research · 1d ago
Amazon launches research initiative with Stanford University to advance AI and science
Amazon Science homepage · Research · 1d ago
Improving synthesis prediction of small molecules at scale with RetroChimera
Microsoft Research · Research · 1d ago
Showing 240 of 308 loaded entries. Search and all-headlines view include all available entries in this section. Browse the feed archive.
Browse the news
Recent stories
Headlines grouped by publisher. Search this page by source or headline; use the archive for older entries.
Spacing
Order
"As a Language Model...": Chat Template Switches LLM Self-Referential Voice and Activation Steering Reproduces It
Abstract page for arXiv paper 2609.25021: "As a Language Model...": Chat Template Switches LLM Self-Referential Voice and Activation Steering Reproduces It
Federating Quantum and Classical Computing: A Privacy-Preserving Hybrid Approach
Abstract page for arXiv paper 2609.25082: Federating Quantum and Classical Computing: A Privacy-Preserving Hybrid Approach
Entropy Can Flow, or It Can Guide. Be Entropy. LEDFlow: Introducing Entropy-guided Generation Order into Uniform Discrete Flow
Abstract page for arXiv paper 2609.25131: Entropy Can Flow, or It Can Guide. Be Entropy. LEDFlow: Introducing Entropy-guided Generation Order into Uniform Discrete Flow
The Probabilistic Structure of Large Language Models
Abstract page for arXiv paper 2609.25134: The Probabilistic Structure of Large Language Models
Stable Unsupervised Continual Chunking with Sheaf SyncMap
Abstract page for arXiv paper 2609.25143: Stable Unsupervised Continual Chunking with Sheaf SyncMap
Brain-Inspired Hierarchical Modularity for General Continual Learning
Abstract page for arXiv paper 2609.25146: Brain-Inspired Hierarchical Modularity for General Continual Learning
Dual-GNN Multilevel Coarsening for Maximum Independent Set
Abstract page for arXiv paper 2609.25149: Dual-GNN Multilevel Coarsening for Maximum Independent Set
Exposing Blind Spots in Deep Imbalanced Regression Evaluation
Abstract page for arXiv paper 2609.25152: Exposing Blind Spots in Deep Imbalanced Regression Evaluation
Learning Neural Feedback Linearization for Data-driven Systems via Augmented Lagrangian
Abstract page for arXiv paper 2609.25163: Learning Neural Feedback Linearization for Data-driven Systems via Augmented Lagrangian
Mitigating Sequential Reappearance in Diffusion Data-Point Unlearning
Abstract page for arXiv paper 2609.25166: Mitigating Sequential Reappearance in Diffusion Data-Point Unlearning
Multi-Term Fourier Graph Neural Network with Sample Relationship Learning for Enhanced Remaining Useful Life Prediction
Abstract page for arXiv paper 2609.25179: Multi-Term Fourier Graph Neural Network with Sample Relationship Learning for Enhanced Remaining Useful Life Prediction
Trains but Doesn't Learn: A Post-Training Delivery Benchmark for LLM Agents as Forward-Deployed Engineers
Abstract page for arXiv paper 2609.25237: Trains but Doesn't Learn: A Post-Training Delivery Benchmark for LLM Agents as Forward-Deployed Engineers
Correcting Within-Group Self-Selection Bias in Prioritized Replay
Abstract page for arXiv paper 2609.25297: Correcting Within-Group Self-Selection Bias in Prioritized Replay
Topological Signal Processing With Unoriented Operators
Abstract page for arXiv paper 2609.25310: Topological Signal Processing With Unoriented Operators
Spatiotemporal Kronecker Covariance Neural Networks
Abstract page for arXiv paper 2609.25326: Spatiotemporal Kronecker Covariance Neural Networks
MT-ProtBERT: Multi-task Learning ProtBERT for Intrinsically Disordered Proteins Classification with Scarce Data
Abstract page for arXiv paper 2609.25334: MT-ProtBERT: Multi-task Learning ProtBERT for Intrinsically Disordered Proteins Classification with Scarce Data
Concept Drift from a Causal Perspective
Abstract page for arXiv paper 2609.25340: Concept Drift from a Causal Perspective
Extending FunctionGemma for Practical On-Device Mobile Function Calling
Abstract page for arXiv paper 2609.25373: Extending FunctionGemma for Practical On-Device Mobile Function Calling
Deep Reinforcement Learning on Item-Compatibility Graphs for One-Dimensional Bin Packing
Abstract page for arXiv paper 2609.25397: Deep Reinforcement Learning on Item-Compatibility Graphs for One-Dimensional Bin Packing
Predictive Uncertainty for Neural CAE Surrogates
Abstract page for arXiv paper 2609.25430: Predictive Uncertainty for Neural CAE Surrogates
What Does Chain-of-Thought Entropy Measure? A Channel Audit of Scaffolding, Routing, and Content
Abstract page for arXiv paper 2609.25039: What Does Chain-of-Thought Entropy Measure? A Channel Audit of Scaffolding, Routing, and Content
FREESIA: Covariance-Aware Posterior Transport for Expressive and Scalable Data Assimilation
Abstract page for arXiv paper 2609.25085: FREESIA: Covariance-Aware Posterior Transport for Expressive and Scalable Data Assimilation
The Probabilistic Structure of Large Language Models
Abstract page for arXiv paper 2609.25134: The Probabilistic Structure of Large Language Models
The Informational Content in Lepto-Variance and Its Relation to Higher Moments
Abstract page for arXiv paper 2609.25144: The Informational Content in Lepto-Variance and Its Relation to Higher Moments
Variational objectives for amortized Bayesian inference in inverse problems: The role of posterior conditioning
Abstract page for arXiv paper 2609.25145: Variational objectives for amortized Bayesian inference in inverse problems: The role of posterior conditioning
Empirical Auditing of Edge-Private Graph Generators
Abstract page for arXiv paper 2609.25155: Empirical Auditing of Edge-Private Graph Generators
Penalized Nonreversible Langevin for Constrained Sampling
Abstract page for arXiv paper 2609.25381: Penalized Nonreversible Langevin for Constrained Sampling
PICPIs: Prediction-Interval-Conditional Prediction Intervals
Abstract page for arXiv paper 2609.25388: PICPIs: Prediction-Interval-Conditional Prediction Intervals
A Practical Recipe for Semi-Supervised Federated ASR: Online Pseudo-Labels with Server Update Stabilization
Abstract page for arXiv paper 2609.25471: A Practical Recipe for Semi-Supervised Federated ASR: Online Pseudo-Labels with Server Update Stabilization
Scalable Minimum-Volume Simplex Estimation with Non-asymptotic Analysis
Abstract page for arXiv paper 2609.25576: Scalable Minimum-Volume Simplex Estimation with Non-asymptotic Analysis
Generalized Deep Regression for Repeated Measurements
Abstract page for arXiv paper 2609.25605: Generalized Deep Regression for Repeated Measurements
On the Gradient Heterogeneity Dynamics of Adversarially Robust Federated Regression
Abstract page for arXiv paper 2609.25705: On the Gradient Heterogeneity Dynamics of Adversarially Robust Federated Regression
Optimal Tradeoffs Between Network Size and Parameter Magnitude in Neural Approximation and Minimax Regression
Abstract page for arXiv paper 2609.25710: Optimal Tradeoffs Between Network Size and Parameter Magnitude in Neural Approximation and Minimax Regression
Statistical Gains from Looped Estimation under Parameter Budgets
Abstract page for arXiv paper 2609.25778: Statistical Gains from Looped Estimation under Parameter Budgets
Conditional Tensor Diffusion: Distributional Counterfactual Learning and Inference
Abstract page for arXiv paper 2609.25924: Conditional Tensor Diffusion: Distributional Counterfactual Learning and Inference
Learning to Fluctuate: Statistical Foundations for Causal Tabular Pretraining
Abstract page for arXiv paper 2609.26290: Learning to Fluctuate: Statistical Foundations for Causal Tabular Pretraining
Error Bounds for Statistical Estimators in BTL Model with Parametric Multivariate Utility Functions
Abstract page for arXiv paper 2609.26326: Error Bounds for Statistical Estimators in BTL Model with Parametric Multivariate Utility Functions
SuperPCA: subspace analysis and an efficient algorithm for high-dimensional PCA
Abstract page for arXiv paper 2609.26406: SuperPCA: subspace analysis and an efficient algorithm for high-dimensional PCA
A Practical Guide on Graphical Model Validation
Abstract page for arXiv paper 2609.26445: A Practical Guide on Graphical Model Validation
On Basis Function Selection for Sparse Gaussian Process Regression
Abstract page for arXiv paper 2609.26624: On Basis Function Selection for Sparse Gaussian Process Regression
Poitras Center to fuel early careers of 50 young scientists dedicated to psychiatric disorders research
Patricia and James Poitras launched a fellowship program for MIT graduate students and postdocs studying major mental illness, expanding their philanthropy to directly support ear…
A new chapter for MIT Reads
The popular MIT Reads program will begin a new focus on fiction and memoir as a way to help the MIT community celebrate the power of storytelling and strengthen social connection.
New AI technique could make minimally invasive surgeries safer and more precise
Researchers designed an AI-driven system that could boost the safety and speed of minimally invasive surgical procedures by rapidly matching X-rays captured during surgery with a…
Measure by measure, studying society accurately
Naoki Egami is an MIT political scientist specializing in the methodology of research, including the question of “external validity” — whether the results of particular studies ap…
MIT spinout turns plastic waste into resilient building materials
Atlas Building Composites, an MIT spinout, developed an AI-powered robotic manufacturing platform for recycling single-use plastics into durable building materials.
New method enables AI for safety-critical situations
HardFlow is a new algorithm developed at MIT that helps pretrained generative AI models satisfy hard constraints while improving solution quality without retraining, in applicatio…
Lifesaving Lincoln Laboratory device wins 2026 Excellence in Technology Transfer Award
AI-GUIDE, an AI-assisted catheterization device developed by MIT Lincoln Laboratory and Massachusetts General Hospital, was chosen for the Federal Laboratory Consortium’s 2026 Exc…
MIT Schwarzman College of Computing launches pilot to help educators teach AI across disciplines
A weeklong summer workshop brought higher-education faculty to MIT's campus to explore how AI and machine learning materials can be adapted for their classrooms.
From MIT to IBM, expediting AI and quantum deployment
MIT affiliates engage with the MIT-IBM Computing Research Lab to bring rigorous theory to production systems in reinforcement learning and AI agents, quantum machine learning, and…
System helps humans predict when self-driving cars will make mistakes
The CW-Net technique explains the behavior of an autonomous vehicle, using concepts a human can easily understand. Researchers found these explanations helped drivers predict how…
Walter Torous named executive director of MIT Center for Real Estate
MIT Senior Lecturer Walter Torous has been appointed executive director of MIT Center for Real Estate. He will lead teaching, fundraising, and events while directing the MSRED pro…
Ila Kumar: Innovating with communities
MIT PhD student Ila Kumar works alongside young people who have gone through the child welfare system, to reimagine how technology can support healing, connection, and independenc…
MIT Quantum Initiative launches postdoctoral fellowship program
The MIT Quantum Initiative has launched a postdoctoral fellowship program supporting interdisciplinary research across quantum computing, sensing, materials, simulation, and netwo…
How an MIT research project became a global programming language
Julia, the programming language that originated as a research project at MIT, has gained a loyal following among scientists, engineers, mathematicians, and others. The free and op…
Looking beyond natural sequences
A machine-learning framework developed by MIT biologists aims to improve the success rate of computational protein design while moving away from results that reproduce sequences f…
AI helps design new materials that work in the real world
MIT researchers added a new component to AI models that design new materials, helping ensure the material will be stable and practical for real-world use. The “CrysVCD” approach c…
Generating scenarios for extreme events, without extreme data
MIT engineers developed a tool that predicts plausible extreme events and worst-case scenarios, such as an extreme storm’s likely duration, intensity, and area of impact. Importan…
Paving the way for greener ammonia production
Ammonia is essential for fertilizer, but its production generates about 1.5 percent of global greenhouse gas emissions. MIT’s Bilge Yildiz and colleagues have developed a computat…
When AI art has no author: Study finds generated images often can’t be traced to training data
Images generated by AI models trained on massive datasets often can’t be traced to specific training images, MIT CSAIL researchers found. Removing individual images from the datas…
Q&A: Rethinking how innovation happens
MIT Professor Eugene Fitzgerald’s book, "The Invisible Engine: Why Innovation Evades Control," explores what innovation really is and what drives value creation. He draws on his o…
Watch scientists decipher burnt scrolls without unrolling them
By making and burning their own papyrus, researchers have come up with a method that might help to read glowing text from unopened scrolls from Herculaneum.
Generative AI designs functional thiolation domains for reprogramming non-ribosomal peptide synthetases
Large language models and generative protein design promise to accelerate biotechnology, but it remains unclear whether they can engineer dynamic megasynth(et)ases whose activity…
Reproducibility in the era of large language models
Reproducibility has long been a concern in machine learning research. The rise of large language models adds new layers of complexity, underscoring the need for clearer reporting…
AI co-scientists are revolutionizing how research is done
Artificial-intelligence systems can generate hypotheses, design experiments and analyse data — but humans still need to decide what makes sense.
Generating protein hydrogels with customizable stress relaxation behavior via deep learning-driven entanglement design
Protein hydrogels are promising artificial extracellular matrices (ECMs) for 3D stem cell and organoid culture due to their favorable stress relaxation behavior (a decrease in str…
Bayesian bilevel operator learning with low-rank adaptation for efficient uncertainty quantification of PDE inverse problems
Uncertainty quantification in PDE inverse problems is essential in many applications. Scientific machine learning and AI enable data-driven learning of model components while pres…
AI cracked the Navier–Stokes challenge. What does that mean for physics?
Physicists and mathematicians are going beyond the classic equations of fluid dynamics to understand turbulence — often with the help of AI.
ConvexGating infers gating strategies from clusters in single cell cytometry data
Manual expert gating remains common practice for defining specific cell populations in flow cytometry data, but increasing numbers of measured parameters and high inter-rater vari…
ReScale4DL: balancing pixel and contextual information for enhanced bioimage segmentation
Deep learning is the state-of-the-art approach for bioimage segmentation. However, it presents a paradox regarding image resolution: counterintuitively, deep learning segmentation…
How fast are you ageing? Ask AI
A new system will help scientists to refine large language models for longevity research and clarify ‘biological’ age.
How a team of AIs discovered a promising lung-cancer drug
Researchers developed a ‘virtual biotech’ made up of as many as 37,000 agents reporting to an 'chief scientist'.
Advanced quantitative mapping of Alzheimer’s disease neuropathology and microglial activation in post-mortem hippocampal tissue
We developed a high-throughput imaging workflow to spatially map Alzheimer’s disease (AD) pathology in postmortem hippocampal and medial temporal lobe sections from 65 University…
Turning scientific research papers into interactive AI agents
A virtual corresponding author prompted in natural language can explain and reproduce a paper’s analyses.
AI companies must work with the research community to protect attribution
An era-defining mathematics claim raises questions over how AI tools credit earlier work.
AI tool turns any paper into an ‘agent’ that can collaborate and answer complex queries
The Paper2Agent system makes it easier for researchers to reproduce papers and understand work in unfamiliar fields, authors say.
Deep learning coupled with scalable domain-specific structural validation expands RNA virus discovery from metatranscriptomes
The discovery of RNA viruses from metatranscriptomic data remains challenging due to extreme sequence divergence and length heterogeneity, ranging from massive polyproteins to sho…
Identification of broadly tumour-reactive γδ TCRs from multiple myeloma
γδ T cells are becoming increasingly appreciated for their antitumour capacity and role in mediating responses to immune checkpoint blockade1–3. Unlike classical αβ T cells, the d…
Exploring the mitochondrial landscape in trabecular meshwork of primary open-angle glaucoma for novel therapeutic targets
Glaucoma ranks among the primary contributors to irreversible vision loss globally, with its pathogenesis closely associated with mitochondrial function. This study aims to identi…
On-demand design of controlled-release systems using an expert-mimic AI framework
Navigating a vast formulation-preparation design space that is highly sensitive to perturbations makes trial-and-error approaches inefficient and costly for diverse clinical needs…
‘Multifunctional’ brain implant translates speech and gestures in real time
A neural device uses artificial intelligence to read brain activity for intended words and gestures simultaneously.
Amazon launches research initiative with Stanford University to advance AI and science
The collaboration aims to advance research while broadening participation and translating discovery into real-world solutions.
Advancing AI for biology: Teaching models to design and characterize antibodies
Three new papers from Amazon Bio Discovery address bottlenecks in AI-driven antibody engineering, from benchmarking binding predictors to experimentally validating de novo design.
Why don’t machine learning research agents overfit?
New research shows that AI agents — like human research communities — learn strategies compressible enough to fit in a few tokens, which prevents memorization and explains why ben…
Developing provably correct Rust code with Verus
Amazon uses the open-source Verus tool to mathematically prove the correctness of critical Rust code, including key components of the Nitro Isolation Engine.
When LLM judges agree, should we believe them?
Discounting the opinions of LLM judges with highly correlated outputs ensures that panels of judges reflect a true diversity of perspectives.
SOP-Bench: A new benchmark for evaluating AI agents on real business procedures
Extendable framework enables testing agents on the full set of capabilities required to successfully complete a procedure, not isolated proxy tasks.
A decade of mathematical certainty: Reflections on the Automated Reasoning Group
Byron Cook reflects on 10 years of using mathematical proof to verify AWS infrastructure, from IAM Access Analyzer to AI guardrails.
AWS Trainium Frontier competition: Co-design models and kernels on purpose-built AI chips
A competition with a finalist ceremony during NeurIPS 2026, challenging researchers to train language models from scratch on Trainium, exploring what optimal architectures look li…
34 Amazon Research Awards Build on Trainium recipients announced
Amazon announces 34 recipients of the Build on Trainium program, a $110 million credit initiative supporting AI research at 30 universities including Stanford, UC Berkeley, UIUC,…
How controllers from industrial machinery can coordinate multitask machine learning
Instead of compromising among parameter updates dictated by different training objectives, ControlG allocates computational capacity to objectives sequentially and dynamically.
A new benchmark for evaluating patient-facing health AI agents
PatientAgentBench generates a synthetic patient health record, a realistic clinical vignette, and a patient agent that converses with the AI system under evaluation, to capture wh…
Amazon is investing in the Lean Focused Research Organization
As AI agents take on higher-stakes decisions, Lean programming language makes it possible to mathematically prove they will behave safely.
Improving synthesis prediction of small molecules at scale with RetroChimera
Custom-made molecules are advancing medicine, materials, and agriculture, but producing them is slow and expensive. A new Nature paper highlights RetroChimera, a predictive model…
Called to serve: Tech, research, and positive impact with Chris White
Lab Director Chris White has worked on research challenges with real-world implications—from new approaches to wartime data analysis to tools for combating human trafficking. He t…
GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models
What if pathology foundation models could do more with less? GigaPath-Flash and GigaTIME-Flash cut computational demands while maintaining strong performance, opening the door to…
Broadening access to Skala creates a faster path to predictive DFT
Skala 1.1, the updated deep-learning exchange-correlation functional from Microsoft Research, provides greater accuracy, expanded accessibility across the computational chemistry…
MindTopo reveals VLMs’ spatial reasoning abilities
A path, a fence, a knot. MindTopo sets a new benchmark for testing how AI understands topological relationships and highlights new opportunities to strengthen spatial reasoning an…
Introducing CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement
Radiology AI is evolving beyond report generation. CARE-X explores a unified approach that combines flexible reasoning, calibrated predictions, and measurement-based tools for che…
Orchard: An open framework for scalable agentic AI
Orchard is an open-source framework for the research community to train and evaluate AI agents across task types. It reduces complexity while supporting strong performance from sm…
Echoverse: Deep, evolving environments for computer-use agents
Computer-use AI agents struggle with multi-step workflows like email and customer support. Echoverse trains agents in realistic environments rather than simply providing more trai…
EvoLib: Turning experience into evolving knowledge
LLMs do not get smarter just by remembering more. EvoLib turns experience into evolving knowledge, taking reusable skills and insights that help models learn and adapt across task…
How Virus-like ‘Jumping Genes’ Became Our Partners in Evolution
Half of our genome is made of transposons — snips of DNA that can move and copy themselves. But they’re more than parasites or genetic junk.
Mathematicians Build Long-Awaited Graph Sandwich
The proof of a decades-old conjecture has given researchers a new way to understand complex networks.
Where Does the Quantum World End and Ours Begin?
Jonathan Halliwell explains how quantum decoherence is key to understanding how we transition from a world with a wave-like nature of matter and energy to the classical macroscopi…
Ctenophores Aren’t Just Beautiful. They’re Biological Wonders.
Comb jellies are helping answer fundamental questions in biology, from how the earliest nervous systems evolved to how bioluminescence works.
Black Holes or Black Hole Stars? Astronomers Spar Over Webb Telescope’s ‘Little Red Dots.’
The James Webb Space Telescope spots mysterious “little red dots” everywhere. A bold new theory suggests they’re suns dozens of times larger than our entire solar system.
Why Do These Fossil Shells Flip Their Spirals Every Few Millennia?
The mystery of the flipping foraminifera may conceal a rarely observed evolutionary process playing out across the planet.
The Four-Color Theorem Gets a Rare New Proof
By revisiting the famous problem — which was controversially solved in the 1970s with the help of computers — mathematicians have gained important new insights into the nature of…
What Is Math’s Mysterious Langlands Program Really About?
Hidden correspondences hint at a deeper structure to the mathematical universe. Our columnist unpacks one of those connections and asks mathematicians what they might mean.
AI Has Solved One of Math’s $1 Million Millennium Prize Problems
Mathematicians at OpenAI showed that the Navier-Stokes equations, which describe how fluids flow, can sometimes “blow up.” But the massive result is not without controversy.
Live from ICM 2026: What Is Math For in the Age of AI?
In this special live recording from the International Congress of Mathematicians in Philadelphia, July 2026, the central question discussed was what do mathematicians really value…
In an Age of AI, a Physicist Seeks What Endures
Sarah Demers, chair of the physics department at Yale University, sees the potential for large language models to both help and harm her field.
Genome Duplication Is a Radical Evolutionary Gamble
By doubling their entire genome, organisms can evolve extremely rapidly — or they can lose it all. New studies reveal how to survive the high-risk, high-reward evolutionary event.
‘Stunning’ Percolation Proof Solves Decades-Old Puzzle About Phase Transitions
Mathematicians found that a broad class of networks will abruptly shift behavior past a critical point.
‘Stunning’ Percolation Proof Solves Decades-Old Puzzle About Phase Transitions
Mathematicians found that a broad class of networks will abruptly shift behavior past a critical point.
Does Computer Science Need Computers?
The theoretical side of the field doesn’t require computing machines. But many questions would never have been posed without them.
In Hilbert Space, All Things Are Quantumly Possible
To explore quantum phenomena, we must leave the familiar world and enter the abstract realm of Hilbert space.
A New Framework for How the Brain Compresses Our Noisy World
An updated view of categorization reflects the modern understanding that the nervous system is more of a prediction engine than a filing cabinet.
‘Huge Breakthrough’ in the Math of Imbalance
For the first time in 30 years, computer scientists have found a better way to allocate objects evenly between two groups.
Are We Thinking Correctly About AI Intelligence?
Computer scientist Melanie Mitchell discusses why artificial intelligence doesn’t “think” or “reason” like humans, and how we can create better methods for measuring machine cogni…
Building a Quantum Computer, One Fragile Qubit at a Time
No one yet knows which technology will power the quantum computers of the future, but the race to create them has already produced some of science’s most intricate machinery.
MilleMiglia: A realistic instance generator for middle-mile logistics
The future of practice: Enabling teachers to create learning interactives with generative UI
Bypassing inference bottlenecks: Accelerating complex AI search with Retrieve-for-Train
ToolGrad: Efficient tool-use dataset generation with textual "gradients"
Transfer learning for genomic prediction in underrepresented populations
A connectomics milestone: Mapping the complete male fruit fly brain
Mapping global methane emissions from space with deep learning
TimesFM-3: A zero-shot foundation model for multivariate forecasting
Planetary prediction engine: Automating global models via Earth AI
GlucoFM: Foundation model for continuous glucose monitoring
AgentHands: Generating interactive hand gestures for spatially grounded agent conversations in XR
An AI tool for prioritizing candidate biomarkers from wearable sensor data
How mobility gives language models a deeper understanding of place
Seeing beyond BMI: Estimating cardiometabolic risk with smartphone imagery
Empty shelves or lost keys? Recall is the bottleneck for parametric factuality
Advancing AMIE towards expert-level audio-visual clinical consultations
Science One Framework: A verifiable autonomous research framework via Chain-of-Evidence
SymptomAI: Towards a conversational AI agent for everyday symptom assessment
Towards a quantum computer that learns from its errors
Dynamically Scaled Activation Steering
Activation steering has emerged as a powerful method for guiding the behavior of generative models towards desired outcomes such as toxicity…
REVERSAL-BENCH: A Reversibility Axis and Reset Oracle for Measuring the Reset-Free RL Cliff
A central goal of autonomous reinforcement learning is continuous policy training without external resets. However, existing paradigms…
DACA-GRPO: Denoising-Aware Credit Assignment for Reinforcement Learning in Diffusion Language Models
Diffusion large language models are a compelling alternative to autoregressive models, yet existing RL methods for diffusion treat all…
Glyph: A Multi-Strategy Agentic System for Column Description and Sensitivity-Ontology Tagging of Enterprise Data Catalogs
Enterprise data lakes accumulate tables faster than human stewards can document or classify them, leaving columns with missing descriptions…
Shared Selective Persistent Memory for Agentic LLM Systems
Agentic LLM systems that generate code through multi-turn tool use face a fundamental context problem: each session starts from zero…
Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation
Discrete flow matching generates text by iteratively transforming noise tokens into coherent language, but may require hundreds of forward…
How Value Induction Reshapes LLM Behaviour
Conversational Large Language Models are post-trained on language that expresses specific behavioural traits, such as curiosity…
DiscoSign: Discourse-Aware Text to Sign Language Gloss Translation
Sign language processing systems have traditionally operated at the sentence level, ignoring critical discourse phenomena fundamental to…
SimpleDesign: A Joint Model for Protein Sequence and Structure Codesign
Proteins are fundamental to biological processes, with their function determined by the complex interplay between the amino acid sequence…
Putting Captions to the Test: Evaluating Video Caption Quality through Multiple-Choice Question Answering
Evaluating video captioning remains a critical challenge for Visual Large Language Models (VLLMs). Existing metrics primarily rely on…
REFACTOR-VLA: Unsupervised Library Learning of Typed Motor Programs
Most current vision-language-action (VLA) models—such as OpenVLA, π0, RT-2, and RDT-1B—are “monolithic.” This means they generate raw motor…
Agent Seer: Synthesizing Scenarios from Specification Understanding
Evaluating AI agents that use external tools requires realistic test scenarios that capture how practitioners compose tools and iterate…
LLMs Are Not (Consistently) Bayesian: Quantifying Internal (In)consistencies of LLMs’ Probabilistic Beliefs
Modern AI systems are being deployed in complex domains such as medicine, science, and law, where there is often not a single correct answer…
From Preferences to Principles: Rubric-Based Alignment for Grounded Knowledge Answers
Designing effective reward signals for open-domain question answering is challenging because high-quality responses must simultaneously…
IDEA Prune: An Integrated Enlarge-and-Prune Pipeline in Generative Language Model Pretraining
Recent advancements in large language models have intensified the need for efficient and deployable models within limited inference budgets…
PROOF-Gen: From Optimized Data to Better Distillation
Supervised fine-tuning on teacher-generated trajectories is the standard first stage for distilling tool-calling capabilities into…
Luce: Relightable Gaussians for 3D Asset Generation
High-fidelity image-to-3D generation requires a 3D representation that captures both geometry and appearance. To support relighting and…
STARFlow2: Bridging Language Models and Normalizing Flows for Unified Multimodal Generation
Unified multimodal models that understand, reason over, and generate interleaved text–image sequences remain structurally fragmented:…
Beyond Visual CoT: Internalized Visual Thinking for Proactive Video Reasoning
Multimodal large language models increasingly use visual chain-of-thought (Visual CoT) to reason about spatial, temporal, and embodied…
Multilingual Knowledge Transfer under Data Constraints via Lexical Interventions
Cross-lingual knowledge transfer is critical for building high-performing multilingual language models for languages with insufficient…
What a crowdsourced game revealed about steering Olmo 3
A crowdsourced game built on Olmo 3 showed how people can exploit unexpected model behaviors to stress-test prosocial AI evaluations—and how open access to a model’s internals can…
Teaching future scientists to interrogate AI tools for scientific discovery
University of Washington students put Ai2’s AutoDiscovery to the test, showing how AI can surface promising scientific leads while making human judgment, domain expertise, and rig…
How Goodfire used Ai2’s open post-training stack to trace unwanted model behavior
Goodfire used Ai2’s fully open post-training stack to predict LLM behavioral changes, trace unwanted model behavior back to individual training examples, and test targeted fixes w…
BenchMIRT: What are LLM benchmarks actually measuring?
BenchMIRT is a new method for auditing LLM benchmarks question by question, revealing which capabilities they actually measure and helping researchers build smaller, more focused,…
The hard parts of AI-assisted science
At an Ai2 event marking our expanded collaboration with Providence Swedish, researchers explored the hardest problems in AI-assisted science: keeping systems steerable, grounded i…
Ai2 and Providence Swedish Cancer Institute partner to advance AI-assisted scientific discovery
Ai2 and Providence Swedish Cancer Institute are expanding their collaboration after AutoDiscovery helped researchers uncover and validate a promising new immune signal in invasive…
How researchers adapted Dolma for better Thai language models
Thai researchers adapted Ai2’s open Dolma toolkit to build Mangosteen, a 47-billion-token Thai corpus that filters low-quality web data while maintaining or improving model perfor…
How a Georgia Tech team used the open Olmo stack to trace social reasoning
A Georgia Tech team used Ai2’s fully open Olmo stack to trace social reasoning back to the training data that shaped it, finding that dialogue-rich, interpersonal writing had an o…
When a model reads a drug's class from its name—not its knowledge
Researchers used Olmo 3 and its open training data to show that models can infer a drug’s class from its name instead of knowing the specific medication, and traced that shortcut…
TutorMoments: Do AI tutors know when to help and when to hold back?
TutorMoments is an open, replay-based evaluation framework that tests whether AI tutors can recognize when to support a student and when to hold back and encourage deeper reasonin…
Ai2 expands collaboration with Hugging Face to accelerate open science
Ai2 is expanding its partnership with Hugging Face to give its growing portfolio of fully open models, datasets, benchmarks, and applications the storage, bandwidth, and integrati…
Tracing distinctive language in AI-written text
Stony Brook researchers used our infini-gram engine to trace distinctive phrases in AI-generated writing back to existing sources, finding that top-selling self-published books on…
The OlmoEarth Platform: Geospatial inference at planetary scale
How we built the OlmoEarth Platform to fine-tune geospatial models and run continent-scale satellite inference while managing massive data pipelines, distributed compute, and auto…
Who gets to understand AI?
Why fully open models and research artifacts are essential to independent scrutiny, broader participation, and continued U.S. scientific leadership in AI.
New AI technique could make minimally invasive surgeries safer and more precise
Researchers designed an AI-driven system that could boost the safety and speed of minimally invasive surgical procedures by rapidly matching X-rays captured during surgery with a…
New method enables AI for safety-critical situations
HardFlow is a new algorithm developed at MIT that helps pretrained generative AI models satisfy hard constraints while improving solution quality without retraining, in applicatio…
MIT Schwarzman College of Computing launches pilot to help educators teach AI across disciplines
A weeklong summer workshop brought higher-education faculty to MIT's campus to explore how AI and machine learning materials can be adapted for their classrooms.
From MIT to IBM, expediting AI and quantum deployment
MIT affiliates engage with the MIT-IBM Computing Research Lab to bring rigorous theory to production systems in reinforcement learning and AI agents, quantum machine learning, and…
System helps humans predict when self-driving cars will make mistakes
The CW-Net technique explains the behavior of an autonomous vehicle, using concepts a human can easily understand. Researchers found these explanations helped drivers predict how…
Walter Torous named executive director of MIT Center for Real Estate
MIT Senior Lecturer Walter Torous has been appointed executive director of MIT Center for Real Estate. He will lead teaching, fundraising, and events while directing the MSRED pro…
Looking beyond natural sequences
A machine-learning framework developed by MIT biologists aims to improve the success rate of computational protein design while moving away from results that reproduce sequences f…
AI helps design new materials that work in the real world
MIT researchers added a new component to AI models that design new materials, helping ensure the material will be stable and practical for real-world use. The “CrysVCD” approach c…
Generating scenarios for extreme events, without extreme data
MIT engineers developed a tool that predicts plausible extreme events and worst-case scenarios, such as an extreme storm’s likely duration, intensity, and area of impact. Importan…
Paving the way for greener ammonia production
Ammonia is essential for fertilizer, but its production generates about 1.5 percent of global greenhouse gas emissions. MIT’s Bilge Yildiz and colleagues have developed a computat…
When AI art has no author: Study finds generated images often can’t be traced to training data
Images generated by AI models trained on massive datasets often can’t be traced to specific training images, MIT CSAIL researchers found. Removing individual images from the datas…
The benefits of medical AI assistance vary based on user expertise
New research found non-experts deferred to AI-based assistance in diagnosing skin cancer, even when it was wrong, while clinicians were more likely to catch AI errors.
Alexander Rakhlin named director of the MIT Statistics and Data Science Center
Alexander ‘Sasha’ Rakhlin, the Distinguished Professor in Data, Systems, and Society, IDSS and Brain and Cognitive Sciences at MIT, has been named the next director of the MIT Sta…
Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking
Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking are our most advanced live dialogue models yet, built for natural conversation.
AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome
Explore AlphaGenome Atlas, a catalogue predicting the molecular effects and AVI scores for 9 billion single-nucleotide variants across the human genome.
Introducing WeatherNext 3, our most advanced and accurate global weather AI model
WeatherNext 3, our most advanced global weather AI model, is now in Search, Gemini, Maps, Google Maps Platform, and Cloud.
Proactive cyber defense for governments and enterprises
The Fairwind Program is a limited access program for governments and trusted partners to use our cyber defense tools.
Introducing Gemini 3.8 Flash and 3.8 Flash Cyber
Gemini 3.8 Flash and 3.8 Flash Cyber deliver next-generation intelligence for agentic workflows and cybersecurity.
Introducing agentic video understanding with Gemini
We’re launching agentic video understanding across our latest Gemini models for improved accuracy and lower costs and token usage.
Gemini Omni 1.1 Flash lets you build with more control
Gemini Omni 1.1 Flash brings a new suite of creative controls and generative video capabilities to developers.
Piloting the world's first double-blind AI evaluations
Building trust in proprietary model benchmarks using cryptographically secure environments
Intelligent transcription with Gemini 3.5 Transcribe
Now you can get more intelligent speech-to-text transcription with Gemini 3.5 Transcribe.
From Atari to EVE Online: Building on 15 Years of AI Research in Games
Google DeepMind partners with game developers to build generalist agents like SIMA 2 and unlock breakthrough gameplay experiences across persistent worlds.
Introducing Gemini 3.7 Flash
Gemini 3.7 Flash is our most intelligent workhorse model yet for coding and agents.
Putting sign language AI into users’ hands
Discover our new sign-language-to-text (SL2T) translation model, bringing ASL dictation to Gboard and Live Transcribe on Pixel 11.
WeatherNext: AI model achieves breakthrough in forecasting cyclones
WeatherNext enables accurate cyclone forecasts that can give an extra day of warning. Now we are open sourcing the model.
Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration
Gemini Robotics ER 2 is a step change in video understanding, tool orchestration, and multi-robot collaboration for robotic applications.
We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
Our newest music generation model, Lyria 3.5, delivers significant advancements across musicality, lyrics, and vocal quality, empowering you to craft richer tracks. We’r…
Gemini Robotics 2 brings whole body intelligence to robots
From feet to fingertips — we are teaching robots intelligent whole-body control, fine dexterity, and teamwork to complete a broad range of complex tasks.
AI and the Protestant Vision of Spirituality, Work, and Technology
Should AIs be people too?
Statement: Anthropic warns of AI self-improvement risks, considers a pause
FLI President on the White House Executive Order
Magnificent Humanity – The Pope’s First Encyclical Concerns AI
White House working group on AI – Statement from FLI’s Anthony Aguirre
Forking-Sequences — Part II: Multi-Horizon Forecast Ensembling with Reduced Volatility
Healthcare Benchmarks Are Only as Good as Their Assumptions
Pre-Training Isn’t Bitter Enough
Teaching Vision-Language Models to Speak Cinema
Introducing ARFBench: A time series question-answering benchmark based on real incidents
From CUDA to MLX: How K-Search Brings Decades of Kernel Expertise to Apple Silicon
The BAIR Blog
Teaching LLMs to Update Beliefs for Efficient Long-Horizon Interaction
The BAIR Blog
Intelligence is Free, Now What? <br> Data Systems for, of, and by Agents
The BAIR Blog
2026 BAIR Graduate Showcase
The BAIR Blog
Adaptive Parallel Reasoning: The Next Paradigm in Efficient Inference Scaling
The BAIR Blog
Gradient-based Planning for World Models at Longer Horizons
The BAIR Blog
How Open Models Are Driving AI Research
NVIDIA open models from Nemotron, Cosmos and BioNeMo are fueling the field's biggest research questions at ICML 2026.
NVIDIA Research Unlocks Advanced Grasping, Smarter Autonomous Driving and Agent Training at Scale
New NVIDIA Research breakthroughs show how training at scale — across gripper types, driving scenarios and virtual worlds — creates AI that generalizes to diverse applications.
NVIDIA Enables the Next Era Of Physical AI Research With Agent Skills For Autonomous Vehicles, Robotics And Vision AI
New physical AI agent skills, powered by NVIDIA Cosmos 3, help researchers accelerate data generation, simulation, policy training and evaluation for autonomous system development.
NVIDIA Research Advances Robotics From Simulation to the Real World
Featured at the International Conference on Robotics and Automation, eight new NVIDIA Research papers show how robots trained in simulation are moving into the real world.
After Orthogonality: Virtue-Ethical Agency and AI Alignment
AGI Is Not Multimodal
Shape, Symmetries, and Structure: The Changing Role of Mathematics in Machine Learning Research
What's Missing From LLM Chatbots: A Sense of Purpose
We Need Positive Visions for AI Grounded in Wellbeing
Financial Market Applications of LLMs
Towards Brain MRI Foundation Models for the Clinic: Findings from the FOMO25 Challenge
1. Motivation
Brain Latent Progression Individual-based spatiotemporal disease progression on 3D Brain MRIs via latent diffusion
This article aims at reviewing a Alzheimer’s spatiotemporal disease progression predictive model called Brain Latent Progression (BrLP). All in all, this is ...
A Survey of popular LLM Evaluation Metrics
Large Language Models (LLMs) are increasingly applied to critical domains such as medical report generation, where accuracy and trust are essential. Evaluati...
Open-Source Large Language Models in Radiology: A Review and Tutorial for Practical Research and Clinical Deployment
Open-Source Large Language Models in Radiology
MemSAM: Taming Segment Anything Model for Echocardiography Video Segmentation
MemSAM
Simplifying Deep Temporal Difference Learning
tl;dr The authors propose PQN, a simplified deep online Q-Learning that uses very small replay buffers. Normalization and parallelized sampling from vectoriz...
No matching sources found.
Showing 240 of 308 loaded entries. Search and all-headlines view include all available entries in this section. Browse the feed archive.