An AI’s Hidden Map Links a One-Dimensional Rule to a 2D World
A language model’s hidden representations helped surface a surprising bridge between one-dimensional and two-dimensional cellular automata.
Artificial Intelligence
How Scattered Light Found a Way to Do AI’s Pairwise Math
A real optical experiment found a way for scattered light to generate pairwise features resembling a core comparison in AI attention.
The Hidden Social Logic AI Misses in a Simple Word: “If”
A tiny word exposes a hidden challenge for language models: using context and speaker intent, not just a sentence’s literal logic.
A New AI Method Tracks Probability Across DNA’s Many Paths
A sampling method recovers lost probability in DNA-to-protein language models, reaching estimates that pruning could not feasibly compute.
The AI-Helped Map That Turns Spin Frustration Into a Field
A geometric change of coordinates recasts frustrated spin chains, revealing an exact mapping—and the limits of one AI-assisted discovery.
To Test a Lockdown, This AI First Rebuilds How a City Spends Its Day
A new AI framework models daily activity choices under shifted lockdown timelines, then compares simulated infection and urban-activity trade-offs.
The Missing No: Why AI's Scientific Ideas Dodge Null Results
A large scientist-led test found AI research ideas rarely propose null results—and revealed why human judgment still matters.
A Robot Learns to Recover by Rewinding Its Mistakes in Simulation
A rewindable robot world lets people correct failures in simulation, then tests whether those lessons transfer to real machines.
The Golden Sheep Puzzle That Exposed AI’s Probability Blind Spot
A three-door probability puzzle reveals why language models can ace routine math yet stumble when intuition, wording, or a wrong hint intervenes.
Fake Citations Were Scattered Through Papers, Not Piled Up
A 111-million-reference audit finds fake citations spread thinly through research, challenging the idea that AI errors stay isolated.
Could Thomas Kuhn’s Philosophy Help an AI Find DNA’s Hidden Signals?
An AI system borrowed Thomas Kuhn’s idea of scientific revolutions to rethink DNA motif search, producing a faster algorithm that still awaits lab validation.
An AI Search Engine Learns to Walk the Library Shelves
A new AI search method navigates document hierarchies, finding evidence quickly while risking the loss of answers in pruned branches.
The Code Got Faster, but the AI’s Energy Bill Got Much Bigger
An energy audit of AI-written code finds that faster programs can still use more power, while genuine savings may require millions of runs.
The AI ‘Secret Languages’ Were Easier to Read Than Expected
A study of AI agents’ proposed coded languages finds a twist: other models could often imitate them, while evidence for autonomous secrecy stays limited.
A Hidden Ring Beneath the Ordinary Words an AI Learns to Predict
A language model learned a hidden eight-state pattern in natural-looking text—and its internal representations traced the pattern’s ring.
The AI Climate Model Counts Growth, Not Just Data-Center Power
A new climate-economy model finds AI’s indirect growth effects may rival data-center emissions—and reshape the case for climate policy.
An AI Attack Trick Reveals What Particle-Physics Checks Can Miss
Borrowing adversarial attacks from machine learning, physicists reveal how tiny, plausible input shifts can unsettle collider-event classifiers.
The Scientific Tasks AI Can Do—and the Ones It Can't See
A 200-task benchmark reveals where AI agents excel in science—and why running valid code is not the same as sound scientific judgment.
A Hidden Probability Map Helps a Language Model Guess What's Next
Researchers reveal how language models track hidden states in unfamiliar symbol streams—and show that editing the map changes predictions.
One Synonym Can Send an AI’s Hidden Explanation Off Course
A close look inside language models reveals how synonym shifts can scramble interpretable features—and why formal checks still stop short.