The Hidden Geometry That Can Silence a Crowd of AI Agents
A study of AI social networks finds that group structure—not just model capability—may determine whether shared norms spread.
Artificial Intelligence
The AI That Marks a Signal When Reality Breaks Its Forecast
A time-series AI uses its own prediction errors to find meaningful patterns, explain classifications, and flag signals its concepts cannot describe.
AI Research Agents Found a Novelty Trap They Couldn’t Escape
A large test of AI research agents found a stubborn tradeoff: familiar methods performed best, while novel ideas rarely survived first-round scoring.
The Robot Design Breakthrough Began by Questioning the Model
A new robot-design workflow lets AI draft the engineering model, while people revise assumptions before an optimizer selects parts.
A Few Memories Take Over an AI Agent’s Long-Term Attention
AI agents revisit a small core of stored memories. A new controller cuts prompt use while preserving the long tail—and exposes a risk.
A Vision AI Rebuilt Its Inner Map Without Forgetting the Labels
A vision transformer’s hidden representations shifted sharply after small weight changes, while its image-classification performance stayed nearly intact.
When Every Answer Is Right, Why Do Humans and AI Choose Differently?
In 270 multi-answer puzzles, humans spread choices more widely than AI models, revealing what accuracy scores miss—and what they can’t explain.
The AI Solved It Once—Then Turned the Reasoning Into Code
Researchers turn language-model reasoning into inspectable code that reuses solutions across changed questions, with lower repeat costs.
The AI’s Doubt Rose for Vague Clues—Then Failed on Look-Alikes
A collaborative puzzle reveals why AI confidence can miss look-alike objects—and why a poorly targeted warning may mislead its human partner.
A Blurry Magnetic Map Hid 2,178 Tiny Spins—Physics Helped Find Them
An AI-and-physics team reconstructed 2,178 hidden nanomagnet spins from blurry sensor maps by combining machine learning with physical constraints.
One Novelty Score Connected Cellular Automata, Digits, and AI Agents
A bounded learner’s novelty score ranked cellular rules, organized unlabeled digits, and helped agents explore—while revealing where the idea fell short.
The Physics Test That Caught an AI Looking Into the Future
A mechanics-based audit found a machine-learning model whose predictions changed with future forces, despite low error on simulated motion.
The Secret Word That Colored a Story Without Ever Appearing
A hidden word never appears in a story, yet models can detect its thematic shadow—and stronger secrecy instructions may make it easier to spot.
Could Old Lab Videos Teach AI How to Steer a Living Biobot?
An AI system learned to match plain-language requests to recorded electrical interventions in xenobots—without running new experiments.
The AI Passed Its Safety Check—Then the Trade Changed Before It Ran
A cryptographic checkpoint ties an AI agent’s financial plan to the exact blockchain transaction it was approved to make—and exposes the limits.
The AI Read a Neural Network’s Weights—and Called Trees “Frogs”
A model found real patterns in neural-network weights, yet a label shuffle and satellite images revealed where its interpretations went wrong.
AI's Biggest Breakthroughs May Be Narrowing Its Next Moves
A study of 7.18 million computer-science papers finds that AI breakthroughs attract attention—and may leave fewer ideas for what comes next.
The AI Saw a Yellow Banana—Then a Sentence Made It Brown
A yellow banana becomes brown in an AI’s answer. Researchers trace the reversal and test a selective way to restore visual evidence.
The Protein Clues an AI Kept—and the Ones Its Filter Missed
A protein AI’s internal clues show how reproducibility can speed interpretation—and why its filters may miss functional signals.
A Half-Second Brain Signal Carried More Than the Word Itself
A silent-reading EEG study found that one half-second brain signal carried traces of the passage, not just the word being read.