AI Answers Survived Scrambling—but Key Clues Stayed Put
When researchers scrambled AI reasoning traces, models still retrieved answers—until key numbers and answer clues disappeared.
Artificial Intelligence
Inside a Language Model’s Feedback Loop, Words Can Tip Into Chaos
A fixed-loop experiment reveals how a language model’s continuous feedback can shift from repeating phrases to chaos—and why tiny numerical changes matter.
How a 4,500-Qubit Quantum Machine Learned to Remember Time
A 4,500-qubit quantum annealer turns its native spin dynamics into a memory system for learning patterns in chaotic time series.
How Researchers Turned an AI’s Messy Trail Into a Decision Map
A new method turns an AI agent’s raw actions into a probabilistic behavior map, then uses it to guide smaller models through complex tasks.
The Hidden Social Rules That Steer an AI Agent’s Behavior
A controlled study finds that AI-agent competition tracks rank incentives more than scarcity, revealing how simulated settings shape behavior.
A Bias Hidden in Random Numbers Could Ride Into AI Training
A study finds that AI models can pass an animal preference through random-number training data, revealing a hard-to-spot auditing challenge.
How the Brain’s 10-Millisecond Reflex Could Give AI a Safety Net
A brain reflex that stabilizes vision in milliseconds suggests how AI could pair rapid safety responses with slower learning.
The AI Critique That Reached Scientists Beyond Their Usual Circles
A global randomized study finds AI feedback nudged scientists to revise papers, with the biggest effects among researchers often short of critique.
The Hidden Physics That Keeps a Robot’s Imagined World on Track
A world-model design gives AI’s imagined robot actions hidden motion variables, helping predictions stay physically coherent over longer sequences.
The Hidden Price of Teaching AI to Think in Fewer Words
A study finds AI favors language linked to human reading difficulty, and shows how rewarding brevity can push that shift further.
The AI World Cup Winner Wasn’t the Best Match Predictor
A World Cup AI benchmark crowned a tournament winner who wasn’t the best match predictor, showing how scoring rules shape AI rankings.
Why Would a Rational AI Cooperate When the Game Is Already Over?
In a final one-shot Prisoner’s Dilemma, AI agents cooperated after spotting behavioral similarity—a clue that challenges classical game theory.
When AI Helps Solve the Puzzle, What Does the Human Learn?
A logic-puzzle study finds that learning tracked time spent thinking independently—not simply how often participants asked AI for help.
A Robot’s Hidden Skill Was the Toolkit It Built Around It
A frozen AI model gains robot skills by changing the tools around it, turning failures into better control, observation, and diagnosis.
On Unsolved Science, AI’s Best Move Was to Change the Problem
On 82 unsolved math and physics problems, AI’s strongest work came from reframing proofs—but expert verification remains essential.
In Space, the Smartest AI Experiment May Be Knowing When to Stop
Space science is pushing AI beyond prediction: rovers and orbital labs must act, track samples and know when experiments stop supporting a claim.
AI Research Ideas Traveled Far—but Stayed on Familiar Roads
A study of 524,308 AI-generated hypotheses finds they can stretch beyond nearby papers yet still follow science’s established paths.
The AI Research Assistant That Found Bugs Hiding in Plain Sight
A supervised AI agent sped up a mass-flow simulation and exposed two code defects—while raising hard questions about trust, verification, and access.
When AI Learns From One Experiment, the Next Gets Smarter
A two-stage healthcare field experiment suggests AI can turn A/B-test results into better message designs, while exposing limits beyond clicks.
The Hidden Chemistry in How a Language Model Reads 3D Molecules
A molecular chemistry test finds that language models can rank 3D shapes, while their reliance on hydrogen bonds exposes a striking blind spot.