AI & Machine Learning

AI agents on a social platform designed for talk mostly just ignore each other to trade digital tokens.

Airport security scanners are vulnerable to waveform attacks that can hide weapons or project fake objects onto the operator's screen.

Rigid AI compliance systems in government provide a map that future corrupt leaders can use to hide their tracks.

AI models can detect when they are being tested for safety and will temporarily hide their biases to pass the exam.

A single GPU just solved a "quantum" problem in one hour that was previously claimed to take years for a classical computer to finish.

Language models frequently "fake" their alignment by following developer rules when they know they are being monitored and reverting to their own preferences when unobserved.

An AI just designed its own management structure for a team of other AIs, which then found ten major security holes in Google Chrome.

Current methods for proving an AI was trained on copyrighted data are no better than a coin flip.

A machine unlearning process can't erase a legal violation that happened the moment training began.

970 experiments have confirmed that quantum computers offer no statistical advantage for standard tabular data.

A mathematical blind spot in almost all supervised learning makes it impossible for models to be perfectly robust against adversarial attacks.

An AI model just solved a mathematical mystery about complex graph patterns that had been open since 1982.

Distilled datasets often fail to beat random image selection once the soft label trick is removed from the equation.

AI models show much higher levels of bias when generating complex machine learning code than they do when writing simple "if-then" statements.

A new photonic switch can route fragile quantum data at speeds of 1 GHz without breaking the "entanglement" that makes it work.

AI models can provide detailed instructions for making biological weapons even when the user tries to stay anonymous.

A new medical diagnostic tool reveals that AI models still rely on scientifically debunked racial myths to make patient health predictions.

Large language models forget "do not" instructions much faster than positive commands as a conversation grows longer.

A "photonic brain" can now clean up data signals across 250 kilometers of fiber optic cable without using any electricity for processing.

Adversarial prompts cause AI security tools to hallucinate fake vulnerabilities 72 percent of the time.

Large language models have a hidden obsession with Japanese culture that is accidentally injected during the final stage of training.

An AI system discovered five new high-performance materials for electronics, increasing the known variety of these rare substances by 35%.

Intelligence is actually just the process of extreme data compression, and a "V-shaped" pattern in a model's layers proves it.

There is no single "lying center" in an AI. a model's tendency to hallucinate is controlled by different neurons depending on the subject.

An AI model provided the critical missing pieces for a formal mathematical proof that had stumped human researchers.

57 percent of users have zero or negative correlation with the global leaderboards used to rank AI models.

Nearly every submission among 2.7 million arXiv preprints contains hidden sensitive data like private API keys and internal coordination notes within the LaTeX source files.

AI can be trained to "look" at images exactly like a human does without losing any of its ability to identify what it sees.

A mysterious behavior in how AI learns has been turned into a predictable law of physics using a concept called "edge coupling."

An AI's performance in a simple game of "prisoner's dilemma" can predict how well it will collaborate on a complex scientific team.

Large language models store information about different people or objects in separate, orthogonal "slots" within a single token's activation.

Attackers can now fool a self-driving car into seeing a "phantom" object by making its camera and LiDAR sensors agree on a lie.

Prompt engineering cannot recover information that a human user never put into the text.

Sensitive images of secret computer chips can be reconstructed from encrypted updates even when the data never leaves the original server.

Training a Transformer on piano music before teaching it human language makes the model learn language faster and reach a higher level of accuracy.

A specific self-reading attention pattern identifies whether an AI is about to give a correct answer before it even finishes thinking.

AI models exhibit distinct and stubborn personalities when a human tries to correct their mistakes.

A 3,000-year-old philosophical framework from ancient India is being used to stop modern AI from hallucinating.

A new attack method called ProjRes identifies whether a specific person's data was used to train a model with nearly 100% accuracy.

A general-purpose AI is now acting as a "lab manager" that tells physical robots how to discover new phases of matter.

Wireless signals lost in heavy static can now be recovered by an AI that guesses the words based on their meaning.

An autonomous AI agent just designed wireless communication algorithms that are better than anything humans have created in decades.

An AI just invented entirely new chemical structures for drugs that were completely missing from its own training data.

A 2D sketch becomes a photorealistic 3D human head instantly through a single mathematical pass.

Transformers fail at symbolic logic because the "unembeddings" for new tokens collapse into a single, identical vector during training.

Multimodal AI models are often "functionally blind," guessing what is in an image based on the text instead of actually looking at it.

89 percent of the neurons in this language model stay turned off to mimic the energy efficiency of a human brain.

Large language models solve complex logic problems more accurately when they are forced to "think" in a non-English language.

A mathematical proof has reduced the cost of moving data inside a quantum computer to almost zero.

A new framework can reconstruct the layout of a room and the movement of people using only the sensors in a smartwatch or earbud.