Researchers from Skoltech and Sberbank’s Center for Practical Artificial Intelligence have proposed a new method, TOHA, for detecting hallucinations in large language models operating in retrieval-augmented generation (RAG) systems. The approach analyzes the topological structure of a model’s attention maps and makes it possible to identify responses that are not supported by the provided context. The method does not require training additional models and uses only a small amount of annotated data for configuration.Researchers from Skoltech and Sberbank’s Center for Practical Artificial Intelligence have proposed a new method, TOHA, for detecting hallucinations in large language models operating in retrieval-augmented generation (RAG) systems. The approach analyzes the topological structure of a model’s attention maps and makes it possible to identify responses that are not supported by the provided context. The method does not require training additional models and uses only a small amount of annotated data for configuration.[#item_full_content]

Computer scientists worldwide have been developing a wide range of artificial intelligence (AI) systems. Some of these systems rely on an individual AI agent, while others consist of multiple interacting agents that exchange information, cooperate and revise each other’s responses or predictions.Computer scientists worldwide have been developing a wide range of artificial intelligence (AI) systems. Some of these systems rely on an individual AI agent, while others consist of multiple interacting agents that exchange information, cooperate and revise each other’s responses or predictions.[#item_full_content]

Modern computational tools let scientists explore huge numbers of possible molecules, materials and chemical reactions. But testing every combination in the lab is slow and costly. So how do researchers choose the best “recipe” for their experiment?Modern computational tools let scientists explore huge numbers of possible molecules, materials and chemical reactions. But testing every combination in the lab is slow and costly. So how do researchers choose the best “recipe” for their experiment?[#item_full_content]

Researchers from the University of Warwick report that how faithfully we can build a “digital twin” of a human brain depends not only on computing power but fundamentally on how much of the living brain we can measure, validate and update over time.Researchers from the University of Warwick report that how faithfully we can build a “digital twin” of a human brain depends not only on computing power but fundamentally on how much of the living brain we can measure, validate and update over time.[#item_full_content]

By combining two complementary clues—camera geometry and visual appearance—researchers at the Institute of Science Tokyo, Japan, developed a new approach for preserving identities across multiple cameras. The method uses epipolar geometry to identify spatially consistent candidate matches and appearance similarity to distinguish among possible identities. The approach can be integrated with existing single-camera tracking systems without requiring environment-specific retraining.By combining two complementary clues—camera geometry and visual appearance—researchers at the Institute of Science Tokyo, Japan, developed a new approach for preserving identities across multiple cameras. The method uses epipolar geometry to identify spatially consistent candidate matches and appearance similarity to distinguish among possible identities. The approach can be integrated with existing single-camera tracking systems without requiring environment-specific retraining.[#item_full_content]

For decades, chess has served as a laboratory for studying intelligence and decision-making. It is well established that today’s chess engines can outperform even the strongest grandmasters. But far less is understood about how their play actually differs.For decades, chess has served as a laboratory for studying intelligence and decision-making. It is well established that today’s chess engines can outperform even the strongest grandmasters. But far less is understood about how their play actually differs.[#item_full_content]

Model checking helps automatically verify whether hardware and software systems satisfy specified requirements. It has become an important formal verification technique, but two major challenges remain: state-space explosion, which limits the size of systems that can be checked, and long verification times.Model checking helps automatically verify whether hardware and software systems satisfy specified requirements. It has become an important formal verification technique, but two major challenges remain: state-space explosion, which limits the size of systems that can be checked, and long verification times.[#item_full_content]

Next-generation science experiments will collect more data than ever—so much so that they’ll surpass the capabilities of current data storage and analysis methods. To help, researchers at the Department of Energy’s SLAC National Accelerator Laboratory developed an AI method to compress large amounts of raw data without losing subtle details critical to scientific discovery. They published the work in Nature Machine Intelligence.Next-generation science experiments will collect more data than ever—so much so that they’ll surpass the capabilities of current data storage and analysis methods. To help, researchers at the Department of Energy’s SLAC National Accelerator Laboratory developed an AI method to compress large amounts of raw data without losing subtle details critical to scientific discovery. They published the work in Nature Machine Intelligence.[#item_full_content]

AI chatbots powered by large language models (LLMs) are widely used as recommender systems. While accelerating access to information, AI can also give factually incorrect responses or propagate societal biases.AI chatbots powered by large language models (LLMs) are widely used as recommender systems. While accelerating access to information, AI can also give factually incorrect responses or propagate societal biases.[#item_full_content]

Experts have shown that the most powerful security defenses in Windows 11 can be dismantled by an attacker who never touches the target computer.Experts have shown that the most powerful security defenses in Windows 11 can be dismantled by an attacker who never touches the target computer.[#item_full_content]