Lost books by ancient philosophers recovered from'unreadable' scrolls Long-lost works of ancient philosophy have been recovered from papyrus scrolls that were scorched by the AD 79 eruption of Mount Vesuvius and thought to be impossible to read. For the first time, researchers have used AI to extract the entire surviving text from super-high-resolution 3D scans of a scroll without unrolling it. The scrolls come from the library of Herculaneum, which was buried along with Pompeii nearly 2000 years ago. Scholars have been trying to read the carbonised scrolls, which resemble lumps of charcoal, since the library was discovered in 1752. Physically unwrapping them risks their destruction and the ink they are written in is mostly indistinguishable from the charred papyri - at least to human eyes.

The scroll was recovered from the library of a luxury Roman villa in Herculaneum, near Naples, that was blasted by heat and buried under ash in AD79. The scroll was recovered from the library of a luxury Roman villa in Herculaneum, near Naples, that was blasted by heat and buried under ash in AD79. The surviving part of an ancient scroll that was burnt to a crisp when Mount Vesuvius erupted nearly 2,000 years ago has been virtually unwrapped and read with help from artificial intelligence. Researchers uncovered 20 columns of previously hidden text covering more than a metre of charred papyrus without physically unrolling the scroll. The age of the scroll, named PHerc 1667, makes it one of the oldest in a collection of hundreds recovered from the library of a luxury Roman villa in Herculaneum that was blasted by heat and buried under ash in the volcanic eruption that destroyed nearby Pompeii in AD79.

Our reporters and editors constantly debate which emerging technologies will define the future. Once a year, we take stock and share some educated guesses with our readers. Here are the advances that we think will drive progress or incite the most change--for better or worse--in the years ahead. Rubrik is the exclusive sponsor of the 10 Breakthrough Technologies 2026 and had no editorial influence on this list. Rubrik is a security and AI operations company that aims to secure and accelerate the world's AI transformation.

We're finally reading the secrets of Herculaneum's lost library A whole library's worth of papyri owned by Julius Caesar's father-in-law were turned to charcoal by the eruption of Vesuvius. Deep within a particle accelerator, theoretical physicist Giorgio Angelotti is hard at work. He sets a black cylinder on a mount, bolts it down, then runs through some safety checks before retreating from the chamber, known as "the hatch". "You have to be sure there's no one in the hatch before you close the door," he says. That's because he is about to blast the sample with a super-powerful beam of X-rays.

Training web agents to navigate complex, real-world websites requires them to master $\textit{subtasks}$ - short-horizon interactions on multiple UI components (e.g., choosing the correct date in a date picker, or scrolling in a container to extract information). We introduce WARC-Bench (Web Archive Benchmark), a novel web navigation benchmark featuring 438 tasks designed to evaluate multimodal AI agents on subtasks. WARC-Bench enables sandboxed interactions with dynamic and realistic webpages using Web ARChive files. We show that WARC-Bench is challenging for leading computer-use models, with the highest observed success rate being 64.8%. To improve open source models on subtask, we explore two common training techniques: supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). Experiments show that SFT models obtain a 48.8% success rate on the benchmark. Training with RLVR over SFT checkpoints, even in data-scarce settings, improves the score to 52.8% on WARC-Bench, outperforming many frontier models. Our analysis concludes that mastering these subtasks is essential for robust web planning and navigation, and is a capability not extensively evaluated by existing benchmarks.

W), which is orders of magnitude larger than current datasets.

This post contains a list of the AI-related seminars that are scheduled to take place between 19 August and 30 September 2025. All events detailed here are free and open for anyone to attend virtually. La-Proteina: Atomistic Protein Generation via Partially Latent Flow Matching Speakers: Kieran Didi (PhD student, Oxford) & Tomas Geffner, PhD (NVIDIA Research) Organised by: ML Protein Engineering Sign up to the mailing list for instructions on how to join (scroll to the end of the page). Somekone – Teaching about AI with an explainable social media simulator Speakers: Henriikka Vartiainen and Matti Tedre (University of Eastern Finland) Organised by: Raspberry PI Sign up here to join. Title to be confirmed Speaker: Oscar Leong (UCLA) Organised by: University of Minnesota Check the website nearer the time for Zoom registration details.

Dr. Sabrina Browne, a Texas-based psychiatrist, explains what flood survivors might experience and how to know when it's time to get professional help. Ever feel like your devices are wearing you down? However, here's the catch: a growing connection exists between technology and mental health, affecting everything from focus to sleep. We're breaking down five buzzworthy terms that explain how our digital habits are shaping the way we think, feel, and function. So, if you've ever wondered why your mind feels foggy or why you can't stop scrolling, this is a judgment-free zone, with smart insights and doable tips to help you reset.

Graphical User Interface (GUI) agents have gained substantial attention due to their impressive capabilities to complete tasks through multiple interactions within GUI environments. However, existing agents primarily focus on enhancing the accuracy of individual actions and often lack effective mechanisms for detecting and recovering from errors. To address these shortcomings, we propose the BacktrackAgent, a robust framework that incorporates a backtracking mechanism to improve task completion efficiency. BacktrackAgent includes verifier, judger, and reflector components as modules for error detection and recovery, while also applying judgment rewards to further enhance the agent's performance. Additionally, we develop a training dataset specifically designed for the backtracking mechanism, which considers the outcome pages after action executions. Experimental results show that BacktrackAgent has achieved performance improvements in both task success rate and step accuracy on Mobile3M and Auto-UI benchmarks. Our data and code will be released upon acceptance.