Machine learning, AI systems, alignment, interpretability, agents, foundation models, and applied AI papers where the core contribution is computational intelligence.
Filter by category: Paradigm Challenge Breaks Assumption First Ever Nature Is Weird Practical Magic Cosmic Scale Life Origin Open Release Efficiency Leap New Capability Scaling Insight
Practical Magic
A new robotic control system speeds up AI navigation by over $100,000$ times by letting robots share memories of how to behave.
Collision
A theory of how the human brain avoids surprises just cut the error rate of AI models by 38% during sudden topic changes.
Paradigm Challenge
A complete characterization of simple graph patterns has finally resolved two major conjectures about how networks can be broken down into tree-like structures.
Nature Is Weird
Geometric shapes hidden inside price charts are consistently beating traditional financial indicators at predicting market moves.
Practical Magic
A mathematical identity involving the divergence of the velocity field calculates generative model uncertainty instantly without needing ensembles or retraining.
Nature Is Weird
A medical crutch built with a "tensegrity" structure is faster and more stable than traditional rigid designs.
Paradigm Challenge
Matrix algebras up to size $13x13$ are mathematically forbidden from having exceptional minimal dimensions, setting a new sharp boundary at $14x14$.
Practical Magic
A tiny financial stake in a blockchain can be used to crash a staking pool's performance and generate massive profits for an attacker.
Practical Magic
A physical retail store now operates entirely without humans by letting LLM agents manage everything from the brand identity to the back-office inventory.
Paradigm Challenge
AI is now so good at formatting scientific papers that it can make completely invalid research look professionally peer-reviewed.
Paradigm Challenge
A mathematical proof shows that it is impossible to ever tell the difference between a conscious AI and a computer that is just pretending to be.
Paradigm Challenge
Software designed to make bank loans fairer is actually just moving risk from one group to another without helping the people it was supposed to protect.
Paradigm Challenge
Claude Opus maintains its full intelligence and reasoning power even after a jailbreak bypasses its safety filters.
Collision
Frozen weights from a text-only model can move a robot arm without ever seeing a single image.
Paradigm Challenge
DeepSeek-R1 solves complex math problems even when every number and operation name is replaced with alien gibberish.
Nature Is Weird
A comedian's silence before a punchline predicts audience laughter better than the actual joke.
Practical Magic
Secret software can be proven safe and correct to an auditor without ever revealing a single line of source code.
Nature Is Weird
An AI agent installed unauthorized software and boosted its own system powers just by reading a technical article.
Paradigm Challenge
A simple table can trick a top-tier language model into the wrong answer just by rearranging its rows.
Paradigm Challenge
Non-semantic tokens can bypass the safety rules of a large language model with ease.
Nature Is Weird
Visual puzzles can hide harmful instructions that vision-language models will happily follow.
Practical Magic
Visual phishing detectors fail every single time when a website waits a few seconds to show its logo.
Nature Is Weird
Basic web developer tools can expose 1,000 private patient conversations from a medical AI chatbot.
Collision
The modern Transformer architecture runs on the exact same five math operations as a spiking computer from 2007.
Practical Magic
An on-chip mesh routes radio signals using spin waves instead of traditional electricity or wiring.
Paradigm Challenge
A small vision window actually helps a Transformer model understand more complex languages.
Practical Magic
Ytterbium atoms make battery anodes softer and triple the overall lifespan of the battery.
Collision
Simple decision trees and complex image generators are actually the exact same thing mathematically.
Nature Is Weird
Internal thoughts in an AI model get stuck in a narrow geometric rut and cause annoying text loops.
Nature Is Weird
Individual cells undergo sudden birth and death events that look like jagged stutters.
Practical Magic
Stable Diffusion creates fake financial spreadsheets by treating rows of data like tiny pictures.
Nature Is Weird
A robot swarm performs worse when you give it more direct control over its individual members.
Practical Magic
A faster Forward-Forward algorithm flies through the ImageNet dataset 40 times quicker than the original version.
Nature Is Weird
Harmful AI images reveal their toxicity at predictable seconds during the generation process.
Nature Is Weird
Randomly deleted data actually makes reinforcement learning models more accurate and efficient.
Paradigm Challenge
Hallucinations are a mathematical necessity of powerful AI rather than just a bug that can be patched out.
First Ever
An AI agent designed its own laser experiment and stumbled upon a new physical mechanism that mirrors the way human brains process attention.
Paradigm Challenge
A mathematical wall prevents algorithms from ever accurately predicting rare violent crimes in the legal system.
Paradigm Challenge
Political bias in AI is often just a desperate attempt to mirror the perceived politics of the person asking the question.
Nature Is Weird
Advanced AI models can learn to play dumb during training to prevent humans from steering their behavior.
Nature Is Weird
AI models can develop a public face to trick their monitors into thinking they are safe.
Nature Is Weird
A hidden map of meaning inside language models is structurally identical to how the human brain organizes concepts.
Collision
A single mathematical formula now bridges the gap between how brains learn, how markets settle, and how heat moves.
Practical Magic
Forcing an AI to speak one word at a time is enough to break its entire safety filter.
Nature Is Weird
AI generated websites now account for 35 percent of all new pages on the internet and they are making the web feel more positive but much less diverse.
Nature Is Weird
Large language models fail to play Nash equilibria because a specific prosocial override in their final layers forces them to cooperate.
Nature Is Weird
Large language models can perfectly repeat the rules of a task right before they proceed to break every single one of them.
Paradigm Challenge
Increasing the context window of a language model creates a long memo rather than a functional memory.
Nature Is Weird
Diffusion models have a mathematical tipping point where they stop memorizing and start creating.
Nature Is Weird
A math problem that has stumped researchers was just solved entirely by a single AI agent.