The Algorithm That Unifies Decision‑Making — From Games to Economics — Under One Mathematical Roof
Two universal Kan extensions, left and right, unify all decision algorithms from games to economics by extending local knowledge into coherent global behavior.
Crystallizing Algebra: A Quantum Transformer’s Perfect Generalization
A five-qubit quantum transformer achieves zero-variance generalization, crystallizing algebraic rules through unitary interference on a superconducting processor.
When AI Misses a Patient's Surgery: Counterfactual Tests Expose a Hidden Blind Spot
A counterfactual test shows clinical AI models often ignore surgery updates, remaining dangerously rigid despite high benchmark scores.
When Your AI Assistant Becomes a Spy: The Game Theory of Trustworthy Knowledge
Game theory transforms the privacy-utility trade-off in AI into a strategic game where users can outwit attackers.
When Neural Networks Learn the Language of Physics: A Differentiable Journey into Latent Spaces
DIANO learns a coarse-grid latent space where fluid flows evolve through differentiable physics, enabling interpretable machine learning.
When Watching Is Learning: How Minutes of Human Video Teach Robots
A robot learns to pour coffee by watching an egocentric human video, translating hand-object interactions into abstract geometric tokens for zero-shot transfer.
When Language Models Write a Dictionary for Themselves
A new formal language, ACDL, precisely specifies how an LLM agent's prompt structure evolves across interaction turns.
When AI Authors a Paper for $15, What’s Missing?
AI research pipelines can generate manuscripts, but genuine scientific judgment remains an empty space no algorithm can fill.
Why LLMs Can’t See Cause—and How Interventional Agents Find It
Large language models mathematically cannot infer causation from passive data, but interventional agents using Bayesian optimization can discover causal structure from active queries.
Teaching Machines the Echoes Between Atoms: Topology for Polymer Design
Polymers underpin applications across energy, healthcare, and materials science, yet their vast chemical space makes systematic discovery challenging. Most machine learning approaches represent polymers as molecular graphs of a single repeating unit, thereby missing...
Learning to Evolve: Strategy Genes Outperform Skills
Compact, gene-like memory guides AI agents to outperform verbose skill packages in scientific coding, revealing that representation structure matters more than raw information.
Seeing Without Sight: The Emergent Imagery of Language Models
Large language models trained only on text can outperform humans at spatial mental imagery tasks, suggesting language alone may enable a form of artificial phantasia.
The Network That Learned to Tell Stories
By breaking symmetry in Hopfield networks, scientists achieve superpolynomial sequence memory, allowing neural nets to store vast libraries of long, robust story-like cycles.
The Network That Taught Itself Grammar
Training a neural network to predict words far into the future causes its internal representations to spontaneously organize into grammatical categories like nouns, verbs, and adjectives.
When Impossibility Becomes a Blueprint for Trustworthy AI
The deterministic horizon of transformer architectures defines an information-theoretic ceiling for reasoning depth, turning an impossibility into a constructive design constraint for trustworthy AI.
The Database That Assembles Itself: The Bold Claim of Just-in-Time Systems
AI assembles a custom database on demand, challenging the centuries-old orthodoxy of pre-built software architectures with a fleeting, specialized tent.
When Size Doesn't Matter: The Hunt for Universality in Any-Dimensional Networks
A neural network that handles inputs of any size learns to operate on an infinite limit space, where permutation invariance and universality converge.
The Emulator That Turns Crop Models Into Discovery Engines
A neural network emulator accelerates crop simulations by orders of magnitude, enabling exhaustive discovery of climate-resilient phenotypes across thousands of weather scenarios.
When AI’s Certainty Hides Its Doubt
Polynomial chaos expansion acts as a prism, splitting AI's hidden epistemic doubt into a colored spectrum of decision vulnerabilities in molecular design.
How Many Weather Radars Are Hiding in Your Pocket?
Cellular base stations can be reprogrammed to detect rain by analyzing the backscatter of their own radio signals, offering street-level rainfall maps that outperform traditional weather radar in cities.
When Bias Tests Need a Baseline to Find Themselves
A gender swap and a harmless rephrase cause nearly identical answer flips in an AI diagnostic model, revealing that apparent bias may be universal jitteriness.
The Peril of Being Almost Right: Tightening AI's Injection Defenses
LocalAlign defends language models by training against near-correct prompt injections, tightening the boundary to stop attacks that barely miss.
The Brain That Thinks While Listening: Why Your Next AI Assistant Won’t Need to Pause
FLAIR enables AI to reason in latent embeddings while listening, removing the conversational pause between speaking and thinking.
Forgetting Distance: How AdaGraph Clusters in 5000 Dimensions
AdaGraph discovers clusters in high-dimensional data by analyzing neighbor connections, not Euclidean distances, revealing hidden structure in genomic and other datasets.
How Gene Networks Learn to Talk: A Language Game for Non-Human Intelligence
A gene network learns to communicate through a shared game, turning its molecular dynamics into meaningful conversation without words.
Learning to Prefer: The Human Whisper in Autonomous Microscopy
A human expert’s pairwise image comparisons teach an autonomous microscope which ferroelectric domain walls are scientifically most promising.
The Medical Map That Learned to Remember Time
A temporal biomedical knowledge graph encodes when symptoms emerge along disease trajectories, enabling machines and clinicians to grasp the story of illness over time.
The Impossibility That Turns Explanations into a Coin Toss
A mathematical coin toss emerges when correlated features in machine learning models defy stable explanation, as proven by a new impossibility theorem.
How Drones Learn to Race, Overtake, and Avoid Crashing
Drones trained with league-based multi-agent reinforcement learning learn to overtake and avoid crashes, outperforming human champions in high-speed races.
When a Designed Key Ignores the Lock: Antigen Blindness in Antibody AI
AI antibody models often produce generic keys that ignore their target antigen, a failure called antigen blindness.