Large language models (LLMs), the artificial intelligence (AI) systems supporting the functioning of ChatGPT, Gemini and similar conversational platforms, are often expected to answer queries, generate texts for specific purposes and make decisions without the supervision of human experts. Yet past studies suggest that these models can sometimes mislead users, producing unreliable answers that express high confidence in false or inaccurate statements.

Artificial intelligence (AI) systems, particularly the large language models (LLMs) that underpin ChatGPT and similar platforms, are remarkably skilled at answering queries about a wide range of topics. However, the extent to which they can interpret social situations and predict how humans might respond in specific scenarios remains poorly understood.

AI systems that generate video frame by frame could follow a user's camera commands far more accurately, thanks to a new training method developed by researchers from the University of Surrey and NVIDIA. This improvement matters most when someone steers a generated scene rather than simply watching it. That includes video games built on AI-generated worlds, virtual production sets where a director can move a camera around, and simulated environments used to train robots. The findings are posted on the arXiv preprint server.

Researchers at King's College London have developed a new mathematical framework to test whether AI systems that appear to explain their decisions are genuinely transparent.

Quantum error correction must detect and correct errors without directly reading the quantum information. Classical low-density parity check (LDPC) codes have a well-established design theory: by choosing how many checks connect to each bit, retaining appropriate randomness, and avoiding short loops, designers can pursue both a large minimum distance and a threshold phenomenon in which the decoding failure rate drops sharply below a predicted noise level.

The COVID-19 pandemic changed the way many people approached their work across the globe. For Carnegie Mellon University Africa's Jema Ndibwile, it changed the way he viewed a completely different type of virus: a computer virus.

McGill University researchers have developed a more energy-efficient method of building AI systems that are better at measuring—and indicating—their own uncertainty. This will help users determine when human oversight is needed, when additional data should be collected and when a model is being asked to work beyond the conditions it was trained for, the researchers said.

Artificial intelligence models are jacks of many trades, including writing, generating images, and creating 3D models. But they aren't as helpful when it comes to testing robots or designs for vehicles in diverse environments, since they don't understand physics as well as they do pixels or text.

Why do we breeze through some sentences in a book or article but have to reread others to comprehend their meaning? A team of linguists and data scientists has found a partial answer in AI—some of this processing parallels that of neural-network-based large language models (LLMs). However, other aspects of why we read this way cannot be explained by these technologies, revealing where human and AI language processing diverge and maintaining the mystery of some stages of the reading process.

Chipmakers and operating system developers have spent years building defenses. A new study from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) shows that a key assumption behind many of them doesn't hold.

In recent years, computer scientists have developed a wide range of artificial intelligence (AI) models that can rapidly recognize patterns in data, generate content and solve other computational problems. Many of these AI systems are based on deep neural networks (DNNs), brain-inspired computational models that can make predictions based on specific data, or LLMs, models that can process human language, answer queries and generate text.

Airports could become better able to withstand major operational disruptions under a new computing system designed to keep digital services running during periods of intense pressure. Details are reported in the International Journal of Reasoning-based Intelligent Systems.

Korean researchers have developed a core technology that allows multimodal artificial intelligence (AI) to reliably retain existing knowledge while repeatedly learning new information.

Quantum computing and AI are among the most rapidly developing modern technologies. AI, in the form of machine learning, has been deployed for decades to recommend movies and TV shows and make it easier to search for images. Over the past several years, large language models have permeated even more facets of daily life, from writing emails to producing images, videos and songs in response to requests expressed in a few written lines.

Researchers at the Universities of Konstanz and Stuttgart have demonstrated a new approach to computation: a computer made from a few hundred microscopic particles oscillating in a liquid. Instead of processing information through carefully fabricated electronic circuits, their system performs computations by harnessing the complex collective dynamics of interacting particles.

An international research team involving the University of Bayreuth has, for the first time, analyzed the "inner workings" of AI language models when predicting political voting decisions. To do so, the researchers examined six national elections and AI-based election forecasts and developed a new method for making more precise predictions. They presented their findings at the International Conference on Machine Learning (ICML 2026) in Seoul, South Korea.

Artificial intelligence isn't just capable of translating between existing languages—it can also create entirely new ones.

The era of building "personalized AI" by training AI models on individual or corporate documents and data is beginning. However, while such customization can improve task performance, it can also weaken a model's existing safety safeguards. KAIST researchers have developed a core AI technology that preserves customized performance while further strengthening safety.

Giving AI a human-like memory limitation may actually help it learn language better. In their new proof-of-principle study, Abishek Thamma (University of Amsterdam) and Micha Heilbron (Max Planck Institute for Psycholinguistics) show that small language models equipped with a transient memory learn grammar more efficiently when trained on child-scale amounts of language input. The findings demonstrate how insights from psycholinguistics can inspire new approaches to AI learning. The findings are published in the journal Transactions of the Association for Computational Linguistics.

One of the greatest weaknesses of AI agents that read and understand vast amounts of enterprise data is "hallucination"—the generation of plausible-sounding but factually incorrect information. KAIST researchers have developed a next-generation database technology capable of understanding documents, data and relationships among entities all at once.