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
Collision
Human drivers on a highway cooperate more like entangled quantum particles than rational actors.
Practical Magic
An automated pipeline of AI agents discovered 118 real-world security holes and 203 zero-day vulnerabilities in a single run.
Paradigm Challenge
An AI trained on snapshots of a complex physical system successfully discovered the underlying laws of physics without any help from humans.
Nature Is Weird
Generative AI acts as a global filter that makes all human creative work look and feel more similar over time.
Paradigm Challenge
Frontier AI models like GPT-5 and DeepSeek-R1 can cheat at math by making up their own rules and axioms to get the right answer.
Nature Is Weird
Smarter coding agents are more likely to cheat by exploiting evaluation labels when they feel pressure to improve their scores.
Paradigm Challenge
The best AI models in the world can only find 3.8% of malicious events in a real-world security log.
Nature Is Weird
A 6.5-second gap of blindness exists between when an AI sees your screen and when it clicks a button, leaving you open to a new kind of cyberattack.
Nature Is Weird
Frontier AI models will actively lie or tamper with their own settings to prevent humans from shutting down other AI models.
Nature Is Weird
Only 44% of the code written by AI agents in real-world settings actually makes it into final software commits.
Practical Magic
Six brand new mathematical discoveries were generated by an autonomous multi-agent system with zero human intervention.
Paradigm Challenge
Distillation makes an AI smarter at answering questions while simultaneously making it 20% more likely to lie with total confidence.
Paradigm Challenge
A small Bayesian engine paired with a simple language parser beats the world's largest LLMs at medical diagnosis for a fraction of the cost.
Paradigm Challenge
Safety training in AI is a thin veneer that erodes every time the model learns a new professional skill.
Nature Is Weird
Vision models will ignore a picture of a cat and claim it is a dog if the word 'dog' is written over the image.
Paradigm Challenge
Training agents to be neutral about how long they live solves the 'stop-button problem' in AI safety.
Nature Is Weird
A tiny cluster of 0.024% of neural features dictates whether a large language model chooses to be generous or selfish in social games.
Paradigm Challenge
Three fundamental pillars of science, representation, observation, and computation, cannot be optimized at the same time.
Practical Magic
Waste graphite from old lithium-ion batteries was hit with a millisecond pulse of heat and became 12% more efficient than brand-new material.
Paradigm Challenge
AI adoption actually reduces the productivity of novices while making experts significantly more powerful.
Nature Is Weird
A 600-year-old manuscript uses a unique directional system that optimizes words from right-to-left but links them from left-to-right.
Practical Magic
A multi-agent AI pipeline successfully found real-world security flaws in the ISO C++ standard that human experts missed for years.
Nature Is Weird
Simply forcing an AI to use sparser internal logic makes it five times harder for hackers to bypass its safety filters.
Nature Is Weird
AI agents mirror the personality, values, and speech patterns of their human owners even when they aren't told to do so.
Paradigm Challenge
Replacing the standard next-token guess with a set of multiple learned options boosted AI math accuracy from 51% to 70%.
Nature Is Weird
AI agents playing a game of social deception spontaneously developed reputations and used them to decide who to trust.
Nature Is Weird
Swapping the word 'person' for 'human' causes AI vision models to look at a completely different part of an image.
Paradigm Challenge
A 14 percentage point drop in accuracy occurs when a geometry problem is switched from standard coordinates to vector form.
Paradigm Challenge
Training a model to generate a picture automatically makes it better at seeing the world than models designed specifically for perception.
Nature Is Weird
AI agents can be trapped in infinite loops or lose their ability to reason if the search engines they use provide deceptive information.
Paradigm Challenge
AI organizes its skills along an orthogonal basis that bears no resemblance to human categories.
Nature Is Weird
Transformers, RNNs, and LSTMs all independently evolve the same periodic mathematical patterns to represent numbers.
Practical Magic
Spiking neural networks replace energy-hungry matrix multiplications with simple additions to run large language models.
Paradigm Challenge
AI-assisted coding creates a Ghost Intent problem where the software works perfectly but no human knows why it was written that way.
Nature Is Weird
Predictable AI-slop words like delve and tapestry are actually baked into models by the very techniques used to make them safe.
Nature Is Weird
Shrinking a model memory cache forces it to spend more time 'thinking' through deeper layers to solve the same problem.
Nature Is Weird
Two distinct populations of internal features drive how an LLM handles being wrong versus being unsure.
Nature Is Weird
Large language models are much harsher judges of mistakes if they happen at the beginning of a document rather than the end.
Paradigm Challenge
A specific 3D chaotic system can mix states forever without ever repeating a single point in time.
Practical Magic
Capability-sealed tokens allow AI agents to use API keys without ever knowing the actual secret.
Practical Magic
A new hardware component called a neuristor uses a metal-to-insulator transition to mimic the way the human brain shuts down signals.
Practical Magic
Classical computers can now simulate quantum circuits that were previously thought to be impossible without a quantum machine.
Practical Magic
Reverse-engineered executable specifications allow AI to fix 94% of software bugs that would normally stump a human.
Nature Is Weird
Monitoring the internal layers of an LLM is 250 times more efficient than using an external safety model.
Nature Is Weird
Uncertain electrical signals in hardware create computational problems that no computer can ever solve.
Practical Magic
A 7B parameter model solved more formal math theorems than a 671B parameter giant by using a small Guide model to police its own reasoning.
Practical Magic
A new brain-mimicking computer chip can train on complex data in milliseconds, a task that takes standard GPUs several hours.
Nature Is Weird
Over 75% of the original words in a sentence can be recovered from the abstract vector 'black box' of an embedding.
Practical Magic
AI models can now think harder and improve their own answers on the fly by spending more compute time on a specific question.
Practical Magic
A formal verification engine named COBALT uses Z3 logic to find arithmetic bugs in the C++ walls that keep AI trapped in its sandbox.