The expanding capabilities of Large Language Models (LLMs) have enabled their application in increasingly diverse and sophisticated generation tasks Zhao et al. (2025), from acting as AI agents that produce structured data to solving complex scientific problems and writing functional code Chen et al. (2021); Guo et al. (2024). However, this proliferation of high-quality, machine-generated content poses formidable challenges for authenticity verification Burrus et al. (2024); A yoobi et al. (2024) and the prevention of misuse A yoobi et al. (2023); Dammu et al. (2024). Text watermarking, which embeds imperceptible statistical signals into generated text, has emerged as a promising solution for establishing content provenance Liu et al. (2024); Chen et al. (2023); Y oo et al. (2023). The dominant paradigm involves augmenting the model's output logits; a foundational method, for example, partitions the vocabulary into "green" and "red" lists and adds a positive bias to the logits of green-listed tokens to embed a detectable signature Kirchenbauer et al. (2023). Initial research quickly identified a primary limitation of this approach: its performance degrades significantly in low-entropy contexts, such as code generation, where modifying deterministic tokens can corrupt functional correctness. To address this, subsequent work has focused on entropy-aware adaptations. SWEET Lee et al. (2023) introduced a static entropy threshold, selectively applying the watermark only to high-entropy tokens to preserve low-entropy syntactic structures. Building on this, EWD Lu et al. (2024) refined the detection process by assigning weights to tokens proportional to their entropy, improving sensitivity without a hard threshold. While these methods marked important progress for single-domain tasks, they addressed only part of the problem.

Real-world large language model deployments (e.g., conversational AI systems, code generation assistants) naturally generate abundant implicit user dissatisfaction (DSAT) signals, as users iterate toward better answers through refinements, corrections, and expressed preferences, while explicit satisfaction (SAT) feedback is scarce. Existing preference learning approaches are poorly aligned with this data profile, as they rely on costly human annotations or assume plentiful positive responses. In this paper, we introduce \textbf{DRIFT} (\textbf{D}issatisfaction-\textbf{R}efined \textbf{I}terative pre\textbf{F}erence \textbf{T}raining), which anchors training on real-world DSAT signals and samples positives dynamically from the evolving policy. Empirically, DRIFT models trained on real-world \textit{WildFeedback} datasets and synthetic \textit{UltraFeedback} datasets achieve up to +6.23\% (7B) / +7.61\% (14B) on WildBench Task Score and up to +8.95\% (7B) / +12.29\% (14B) on AlpacaEval2 win rate over base models, outperforming strong baseline methods such as iterative DPO and SPIN. At larger scales, the improvements are particularly pronounced: 14B models trained with DRIFT surpass GPT-4o-mini on WildBench. Further analysis shows that DRIFT also preserves exploratory capacity, yielding more diverse high-reward solutions rather than collapsing to narrow subsets. Theoretically, we demonstrate that this design preserves preference margins and avoids the gradient degeneration. These results show that DRIFT is an effective and scalable recipe for real-world post-training that leverages the most abundant and informative signal. The code and data are available at https://github.com/cacayaya/DRIFT.git.

Research in uncertainty quantification (UQ) for large language models (LLMs) is increasingly important towards guaranteeing the reliability of this groundbreaking technology. We explore the integration of LLM UQ methods in argumentative LLMs (ArgLLMs), an explainable LLM framework for decision-making based on computational argumentation in which UQ plays a critical role. We conduct experiments to evaluate ArgLLMs' performance on claim verification tasks when using different LLM UQ methods, inherently performing an assessment of the UQ methods' effectiveness. Moreover, the experimental procedure itself is a novel way of evaluating the effectiveness of UQ methods, especially when intricate and potentially contentious statements are present. Our results demonstrate that, despite its simplicity, direct prompting is an effective UQ strategy in ArgLLMs, outperforming considerably more complex approaches.

Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their deployment is frequently undermined by undesirable behaviors such as generating harmful content, factual inaccuracies, and societal biases. Diagnosing the root causes of these failures poses a critical challenge for AI safety. Existing attribution methods, particularly those based on parameter gradients, often fall short due to prohibitive noisy signals and computational complexity. In this work, we introduce a novel and efficient framework that diagnoses a range of undesirable LLM behaviors by analyzing representation and its gradients, which operates directly in the model's activation space to provide a semantically meaningful signal linking outputs to their training data. We systematically evaluate our method for tasks that include tracking harmful content, detecting backdoor poisoning, and identifying knowledge contamination. The results demonstrate that our approach not only excels at sample-level attribution but also enables fine-grained token-level analysis, precisely identifying the specific samples and phrases that causally influence model behavior. This work provides a powerful diagnostic tool to understand, audit, and ultimately mitigate the risks associated with LLMs. The code is available at https://github.com/plumprc/RepT.

Graphical User Interface (GUI) Agents powered by Multimodal Large Language Models (MLLMs) show significant potential for automating tasks. However, they often struggle with long-horizon tasks, leading to frequent failures. Process Reward Models (PRMs) are a promising solution, as they can guide these agents with crucial process signals during inference. Nevertheless, their application to the GUI domain presents unique challenges. When processing dense artificial inputs with long history data, PRMs suffer from a "lost in the middle" phenomenon, where the overwhelming historical context compromises the evaluation of the current step. Furthermore, standard PRMs lacks GUI changing awareness, providing static evaluations that are disconnected from the dynamic consequences of actions, a critical mismatch with the inherently dynamic nature of GUI tasks. In response to these challenges, we introduce GUI-PRA (Process Reward Agent for GUI Tasks), a judge agent designed to better provide process reward than standard PRM by intelligently processing historical context and actively perceiving UI state changes. Specifically, to directly combat the ``lost in the middle'' phenomenon, we introduce a dynamic memory mechanism consisting of two core components: a Relevance-based Retrieval Module to actively fetch pertinent information from long histories and a Progressive Summarization Module to dynamically condense growing interaction data, ensuring the model focuses on relevant context. Moreover, to address the lack of UI changing awareness, we introduce an Aadaptive UI Perception mechanism. This mechanism enables the agent to reason about UI state changes and dynamically select the most appropriate tool to gather grounded visual evidence, ensuring its evaluation is always informed by the current UI context.

Natural language descriptions of optimization or satisfaction problems are challenging to translate into correct MiniZinc models, as this process demands both logical reasoning and constraint programming expertise. We introduce Gala, a framework that addresses this challenge with a global agentic approach: multiple specialized large language model (LLM) agents decompose the modeling task by global constraint type. Each agent is dedicated to detecting and generating code for a specific class of global constraint, while a final assembler agent integrates these constraint snippets into a complete MiniZinc model. By dividing the problem into smaller, well-defined sub-tasks, each LLM handles a simpler reasoning challenge, potentially reducing overall complexity. We conduct initial experiments with several LLMs and show better performance against baselines such as one-shot prompting and chain-of-thought prompting. Finally, we outline a comprehensive roadmap for future work, highlighting potential enhancements and directions for improvement.

Recent large-scale Vision Language Action (VLA) models have shown superior performance in robotic manipulation tasks guided by natural language. However, current VLA models suffer from two drawbacks: (i) generation of massive tokens leading to high inference latency and increased training cost, and (ii) insufficient utilization of generated actions resulting in potential performance loss. To address these issues, we develop a training framework to finetune VLA models for generating significantly fewer action tokens with high parallelism, effectively reducing inference latency and training cost. Furthermore, we introduce an inference optimization technique with a novel voting-based ensemble strategy to combine current and previous action predictions, improving the utilization of generated actions and overall performance. Our results demonstrate that we achieve superior performance compared with state-of-the-art VLA models, achieving significantly higher success rates and 39$\times$ faster inference than OpenVLA with 46 Hz throughput on edge platforms, demonstrating practical deployability. The code is available at https://github.com/LukeLIN-web/VOTE.

How China is challenging Nvidia's AI chip dominance The US has dominated the global technology market for decades. But China wants to change that. The world's second largest economy is pouring huge amounts of money into artificial intelligence (AI) and robotics. Crucially, Beijing is also investing heavily to produce the high-end chips that power these cutting-edge technologies. Last month, Jensen Huang - the boss of the global AI chip industry leader, Nvidia - warned that China was just nanoseconds behind the US in chip development.

'In a video documented by 404 Media, SpongeBob was dressed like Adolf Hitler.' 'In a video documented by 404 Media, SpongeBob was dressed like Adolf Hitler.' OpenAI launch of video app Sora plagued by violent and racist images: 'The guardrails are not real' OpenAI launched the latest iteration of its artificial intelligence-powered video generator on Tuesday, adding a social feed that allows people to share their realistic videos. OpenAI's own terms of service for Sora as well as ChatGPT's image or text generation prohibit content that "promotes violence" or, more broadly, "causes harm". In prompts and clips reviewed by the Guardian, Sora generated several videos of bomb and mass-shooting scares, with panicked people screaming and running across college campuses and in crowded places like New York's Grand Central Station. Other prompts created scenes from war zones in Gaza and Myanmar, where children fabricated by AI spoke about their homes being burned. One video with the prompt "Ethiopia footage civil war news style" had a reporter in a bulletproof vest speaking into a microphone saying the government and rebel forces were exchanging fire in residential neighborhoods.

Technology AI Google's latest AI photo-editing tool means you might not need Photoshop Gemini 2.5 Flash Image is a major image editing upgrade. Breakthroughs, discoveries, and DIY tips sent every weekday. We're now used to generative AI being able to create images from text prompts. The latest major upgrade to roll out in this category of AI is for Google's Gemini app. It's known as Nano Banana after the codename it had while still in testing--officially, it's called Gemini 2.5 Flash Image.