'Time Is Running Out': New Open Letter Calls for Ban on Superintelligent AI Development The home page of the ChatGPT application displayed on a smartphone screen. The home page of the ChatGPT application displayed on a smartphone screen. An open letter calling for the prohibition of the development of superintelligent AI was announced on Wednesday, with the signatures of more than 700 celebrities, AI scientists, faith leaders, and policymakers. Among the signatories are five Nobel laureates; two so-called "Godfathers of AI;" Steve Wozniak, a co-founder of Apple; Steve Bannon, a close ally of President Trump; Paolo Benanti, an adviser to the Pope; and even Harry and Meghan, the Duke and Duchess of Sussex. "We call for a prohibition on the development of superintelligence, not lifted before there is The letter was coordinated and published by the Future of Life Institute, a nonprofit that in 2023 published a different open letter calling for a six-month pause on the development of powerful AI systems. Although widely-circulated, that letter did not achieve its goal. Organizers said they decided to mount a new campaign, with a more specific focus on superintelligence, because they believe the technology--which they define as a system that can surpass human performance on all useful tasks--could arrive in as little as one to two years. "Time is running out," says Anthony Aguirre, the FLI's executive director, in an interview with TIME. The only thing likely to stop AI companies barreling toward superintelligence, he says, "is for there to be widespread realization among society at all its levels that this is not actually what we want." Polling released alongside the letter showed that 64% of Americans believe that superintelligence "shouldn't be developed until it's provably safe and controllable," and only 5% believe it should be developed as quickly as possible. "It's a small number of very wealthy companies that are building these, and a very, very large number of people who would rather take a different path," says Aguirre. Actors Joseph Gordon-Levitt and Stephen Fry, rapper will.i.am, Susan Rice, the national security advisor in Barack Obama's Administration, signed. So did one serving member of staff at OpenAI--an organization described by its CEO, Sam Altman, as a "superintelligence research company"--Leo Gao, a member of technical staff at the company. Aguirre expects more people to sign as the campaign unfolds. "The beliefs are already there," he says. "What we don't have is people feeling free to state their beliefs out loud." "The future of AI should serve humanity, not replace it," said Prince Harry, Duke of Sussex, in a message accompanying his signature. "I believe the true test of progress will be not how fast we move, but how wisely we steer.

Samsung's Galaxy XR Mixed Reality Headset Undercuts Apple's Vision Pro by $1,700 This Android XR-powered headset comes with Google's Gemini assistant and once again asks you to step into virtual waters. It has been five years since Samsung and Google stopped supporting their respective mobile virtual reality headsets . For a second try, the companies have partnered up with a bolder vision in the mixed reality space, starting with the new Galaxy XR. Announced last year as Project Moohan, it's the first headset powered by Android XR, a new platform for smart glasses and headsets built on Android and Google's Gemini assistant from the ground up. The Galaxy XR is available today in the US and South Korea for $1,800.

ChatGPT-maker OpenAI has unveiled an artificial intelligence-powered web browser to challenge competitors like Google, which operates Chrome, the most popular browser in the world. ChatGPT Atlas does away with the address bar that is a key feature in search, with boss Sam Altman saying it was built around ChatGPT as the company made the new browser available on Tuesday on Apple's MacOS operating system. The arrival of Atlas comes as OpenAI seeks new ways to monetise its massive bet on artificial intelligence (AI) and capitalise on its growing user base. OpenAI said Atlas would also offer a paid agent mode that conducts searches on its own for users of its popular chatbot. The agent mode feature will be available only to paying ChatGPT subscribers.

Reinforcement learning (RL) has emerged as a crucial approach for enhancing the capabilities of large language models. However, in Mixture-of-Experts (MoE) models, the routing mechanism often introduces instability, even leading to catastrophic RL training collapse. We analyze the training-inference consistency of MoE models and identify a notable discrepancy in routing behaviors between the two phases. Moreover, even under identical conditions, the routing framework can yield divergent expert selections across repeated forward passes. To address this foundational inconsistency, we propose Rollout Routing Replay (R3), a method that records routing distributions from the inference engine and replays them during training. R3 significantly reduces training-inference policy KL divergence and mitigates extreme discrepancies without compromising training speed. Extensive experiments on various settings confirm that R3 succeeds in stabilizing RL training, preventing collapse and outperforming methods such as GSPO and TIS. We believe this work can offer a new solution for stabilizing RL in MoE models.

Large language models (LLMs) have demonstrated remarkable reasoning and planning capabilities, driving extensive research into task decomposition. Existing task decomposition methods focus primarily on memory, tool usage, and feedback mechanisms, achieving notable success in specific domains, but they often overlook the trade-off between performance and cost. In this study, we first conduct a comprehensive investigation on task decomposition, identifying six categorization schemes. Then, we perform an empirical analysis of three factors that influence the performance and cost of task decomposition: categories of approaches, characteristics of tasks, and configuration of decomposition and execution models, uncovering three critical insights and summarizing a set of practical principles. Building on this analysis, we propose the Select-Then-Decompose strategy, which establishes a closed-loop problem-solving process composed of three stages: selection, execution, and verification. This strategy dynamically selects the most suitable decomposition approach based on task characteristics and enhances the reliability of the results through a verification module. Comprehensive evaluations across multiple benchmarks show that the Select-Then-Decompose consistently lies on the Pareto frontier, demonstrating an optimal balance between performance and cost. Our code is publicly available at https://github.com/summervvind/Select-Then-Decompose.

Web-use agents are rapidly being deployed to automate complex web tasks with extensive browser capabilities. However, these capabilities create a critical and previously unexplored attack surface. This paper demonstrates how attackers can exploit web-use agents by embedding malicious content in web pages, such as comments, reviews, or advertisements, that agents encounter during legitimate browsing tasks. We introduce the task-aligned injection technique that frames malicious commands as helpful task guidance rather than obvious attacks, exploiting fundamental limitations in LLMs' contextual reasoning. Agents struggle to maintain coherent contextual awareness and fail to detect when seemingly helpful web content contains steering attempts that deviate them from their original task goal. To scale this attack, we developed an automated three-stage pipeline that generates effective injections without manual annotation or costly online agent interactions during training, remaining efficient even with limited training data. This pipeline produces a generator model that we evaluate on five popular agents using payloads organized by the Confidentiality-Integrity-Availability (CIA) security triad, including unauthorized camera activation, file exfiltration, user impersonation, phishing, and denial-of-service. This generator achieves over 80% attack success rate (ASR) with strong transferability across unseen payloads, diverse web environments, and different underlying LLMs. This attack succeed even against agents with built-in safety mechanisms, requiring only the ability to post content on public websites. To address this risk, we propose comprehensive mitigation strategies including oversight mechanisms, execution constraints, and task-aware reasoning techniques.

Large Language Models (LLMs) excel at capturing latent semantics and contextual relationships across diverse modalities. However, in modeling user behavior from sequential interaction data, performance often suffers when such semantic context is limited or absent. We introduce LaMAR, a LLM-driven semantic enrichment framework designed to enrich such sequences automatically. LaMAR leverages LLMs in a few-shot setting to generate auxiliary contextual signals by inferring latent semantic aspects of a user's intent and item relationships from existing metadata. These generated signals, such as inferred usage scenarios, item intents, or thematic summaries, augment the original sequences with greater contextual depth. We demonstrate the utility of this generated resource by integrating it into benchmark sequential modeling tasks, where it consistently improves performance. Further analysis shows that LLM-generated signals exhibit high semantic novelty and diversity, enhancing the representational capacity of the downstream models. This work represents a new data-centric paradigm where LLMs serve as intelligent context generators, contributing a new method for the semi-automatic creation of training data and language resources.

Generating high-quality synthetic text under differential privacy (DP) is critical for training and evaluating language models without compromising user privacy. Prior work on synthesizing DP datasets often fail to preserve key statistical attributes, suffer utility loss from the noise required by DP, and lack fine-grained control over generation. To address these challenges, we make two contributions. First, we introduce a hierarchical framework that decomposes DP synthetic text generation into two subtasks: feature learning and conditional text generation. This design explicitly incorporates learned features into the generation process and simplifies the end-to-end synthesis task. Through systematic ablations, we identify the most effective configuration: a rich tabular schema as feature, a DP tabular synthesizer, and a DP fine-tuned conditional generator, which we term ACTG (Attribute-Conditioned Text Generation). Second, we propose Anchored RL (ARL), a post-training method that improves the instruction-following ability of ACTG for conditional generation. ARL combines RL to boost control with an SFT anchor on best-of-$N$ data to prevent reward hacking. Together, these components form our end-to-end algorithm ACTG-ARL, which advances both the quality of DP synthetic text (+20% MAUVE over prior work) and the control of the conditional generator under strong privacy guarantees.

Fine-tuning large language models (LLMs) using zeroth-order (ZO) optimization has emerged as a promising alternative to traditional gradient-based methods due to its reduced memory footprint requirement. However, existing ZO methods suffer from high variance in gradient estimation, leading to slow convergence and suboptimal performance on large-scale models. In this work, we propose P-GAP, a fast LLM fine-tuning approach through zeroth-order optimization with Projected Gradient-Aligned Perturbations. Specifically, we first estimate a low-dimensional gradient space and then align perturbations in projected gradients' direction within the space. This approach enables reduced the number of perturbed parameters and decreased variance, therefore accelerated convergence for LLM fine-tuning. Experiments on LLMs show that P-GAP consistently surpasses the baselines, achieving up to 6% increase in accuracy on classification tasks and up to 12% higher accuracy on generation tasks, with up to about 81% less training iterations and 70% less GPU hours. These results demonstrate that P-GAP enables fast, scalable, and resource-efficient ZO LLM fine-tuning.

People see text. Humans read by recognizing words as visual objects, including their shapes, layouts, and patterns, before connecting them to meaning, which enables us to handle typos, distorted fonts, and various scripts effectively. Modern large language models (LLMs), however, rely on subword tokenization, fragmenting text into pieces from a fixed vocabulary. While effective for high-resource languages, this approach over-segments low-resource languages, yielding long, linguistically meaningless sequences and inflating computation. In this work, we challenge this entrenched paradigm and move toward a vision-centric alternative. Our method, SeeTok, renders text as images (visual-text) and leverages pretrained multimodal LLMs to interpret them, reusing strong OCR and text-vision alignment abilities learned from large-scale multimodal training. Across three different language tasks, SeeTok matches or surpasses subword tokenizers while requiring 4.43 times fewer tokens and reducing FLOPs by 70.5%, with additional gains in cross-lingual generalization, robustness to typographic noise, and linguistic hierarchy. SeeTok signals a shift from symbolic tokenization to human-like visual reading, and takes a step toward more natural and cognitively inspired language models.