NO/FOMO

Independent AI signal, once a day

The AI briefing worth opening.

ISSUE DATE2026-02-27ENGLISH EDITION
This issue
—
All time
—

Hacker News

6 stories
01

Experts sound alarm after ChatGPT Health fails to recognise medical emergencies

Experts have raised significant concerns after a version of ChatGPT tailored for healthcare, dubbed 'ChatGPT Health,' reportedly failed to accurately identify and respond to critical medical emergencies. This incident has prompted an alarm among professionals regarding the practical application and reliability of artificial intelligence in high-stakes environments such as patient care. The failure highlights the complex challenges in developing AI systems that can robustly and ethically operate in sensitive domains where human lives are directly impacted. It underscores the urgent need for more rigorous testing, advanced contextual understanding, and robust regulatory frameworks for AI technologies before their widespread deployment in healthcare. The findings are expected to fuel further debate on AI safety, accountability, and the essential role of human oversight in AI-driven medical solutions.

ChatGPTMedical AIAI SafetyHealthcare TechnologyAI ReliabilityArtificial IntelligenceLarge Language ModelNatural Language Processing
02

Anthropic refuses to bend to Pentagon on AI safeguards as dispute nears deadline

Anthropic, a leading artificial intelligence developer renowned for its advanced large language models, finds itself in a notable dispute with the Pentagon concerning crucial AI safeguards, with a looming deadline adding urgency to the situation. The core of the disagreement revolves around the specific safety measures and ethical protocols that Anthropic is being asked to implement for its AI systems, potentially for integration into defense-related applications. This high-profile standoff illuminates the inherent friction between national security priorities and the foundational ethical principles championed by AI development firms. Anthropic's steadfast refusal to yield to the Pentagon's demands signals a broader industry commitment to ensuring responsible AI deployment and mitigating potential risks associated with powerful AI technologies, especially in sensitive operational environments. The resolution of this ongoing dispute is anticipated to significantly influence future frameworks for collaboration between cutting-edge AI innovators and governmental bodies, particularly in establishing robust governance and stringent safety standards for sophisticated artificial intelligence systems.

Artificial IntelligenceAI SafeguardsAI EthicsNational SecurityAnthropicArtificial IntelligenceLarge Language ModelGenerative AI
03

We Built Secure, Scalable Agent Sandbox Infrastructure

This post details the creation of a secure and scalable infrastructure specifically designed for sandboxing AI agents. The primary objective is to enable the safe and efficient operation of autonomous agents by isolating them within controlled environments. This infrastructure addresses critical challenges such as preventing unauthorized resource access, mitigating security vulnerabilities, and ensuring operational stability. By implementing robust sandboxing mechanisms, the system aims to protect the host environment from potentially malicious or errant agent behavior, a crucial aspect for the reliable deployment of agent-based applications. The discussion likely delves into various architectural patterns and technical methodologies for achieving both strong security isolation and high scalability, essential for managing a growing number of agents concurrently. The underlying principles would involve resource management, secure inter-process communication, and robust error handling to guarantee the integrity and performance of agent operations, as hinted by the exploration of "two ways to sandbox agents" on browser-use.com.

AI AgentSandboxingSecurity InfrastructureScalabilityAgent SystemsAI AgentArtificial IntelligenceMachine Learning
04

Show HN: Badge that shows how well your codebase fits in an LLM's context window

The project "Repo Tokens" introduces a GitHub Action designed to quantify a codebase's size in tokens and visualize its fit within a Large Language Model's context window. This tool addresses the emerging benefit of compact codebases, enabling coding agents to process an entire project contextually. Utilizing `tiktoken`, Repo Tokens calculates token counts and dynamically updates a badge in the repository's README. The badge's color indicates the percentage of an LLM's context window the codebase occupies: green for under 30%, yellow for 50-70%, and red for over 70%. The context window size is configurable, defaulting to 200k tokens, aligning with Claude models. This composite action, involving `tiktoken` installation and inline Python execution, updates the README without committing, allowing workflow control over git strategy. The initiative aims to establish token size as a prominent metric, encouraging leaner, agent-friendly codebases.

GitHub ActionCodebase sizeTokenizationLarge Language ModelContext windowLarge Language ModelAI AgentArtificial Intelligence
05

Breaking Free

The story "Breaking Free," originating from a consumer advocacy organization (forbrukerradet.no), likely addresses the critical imperative for individuals to reclaim autonomy and control within the increasingly complex digital landscape. It presumably champions the cause of digital rights, urging consumers to challenge prevalent issues such as pervasive data collection, opaque algorithmic decision-making, and vendor lock-in that restrict personal freedom and choice. The narrative suggests a call to action against systems that exploit user data or manipulate behavior through sophisticated, often AI-driven, processes. It underscores the importance of fostering an environment where consumers possess genuine ownership over their digital identities and data, advocating for greater transparency in AI applications and robust protective measures. This initiative aims to empower users to navigate technology more consciously, ensuring that digital innovations serve human well-being rather than entrapping individuals within restrictive or exploitative ecosystems, ultimately promoting a more equitable and user-centric digital future.

Digital rightsData privacyAlgorithmic transparencyConsumer protectionEthical AIArtificial IntelligenceMachine LearningAI Agent
06

We gave terabytes of CI logs to an LLM

A recent experiment detailed by Mendral involved feeding terabytes of Continuous Integration (CI) logs to a Large Language Model (LLM) to explore its capabilities in processing and extracting actionable insights from vast, unstructured operational data. The initiative aimed to leverage advanced AI for improved system monitoring, anomaly detection, and debugging processes within development pipelines. Significantly, the findings suggest a strong aptitude of LLMs in understanding and potentially generating SQL queries, indicating their potential to interpret complex log patterns or facilitate SQL-based data querying from these logs. This application underscores a promising avenue for automating the analysis of developer workflows and system behavior, transforming raw log data into structured, understandable intelligence. The study highlights LLMs as a powerful tool for operational analytics, offering a new approach to managing and interpreting the deluge of information generated by modern software development environments.

Large Language ModelCI/CDLog AnalysisSQLOperational AnalyticsArtificial IntelligenceLarge Language ModelNatural Language Processing

huggingface

6 stories
01

The Trinity of Consistency as a Defining Principle for General World Models

The construction of World Models capable of learning, simulating, and reasoning about objective physical laws constitutes a foundational challenge in the pursuit of Artificial General Intelligence. Recent advancements represented by video generation models like Sora have demonstrated the potential of data-driven scaling laws to approximate physical dynamics, while the emerging Unified Multimodal Model (UMM) offers a promising architectural paradigm for integrating perception, language, and reasoning. Despite these advances, the field still lacks a principled theoretical framework that defines the essential properties requisite for a General World Model. In this paper, we propose that a World Model must be grounded in the Trinity of Consistency: Modal Consistency as the semantic interface, Spatial Consistency as the geometric basis, and Temporal Consistency as the causal engine. Through this tripartite lens, we systematically review the evolution of multimodal learning, revealing a trajectory from loosely coupled specialized modules toward unified architectures that enable the synergistic emergence of internal world simulators. To complement this conceptual framework, we introduce CoW-Bench, a benchmark centered on multi-frame reasoning and generation scenarios. CoW-Bench evaluates both video generation models and UMMs under a unified evaluation protocol. Our work establishes a principled pathway toward general world models, clarifying both the limitations of current systems and the architectural requirements for future progress.

World ModelsArtificial General IntelligenceMultimodal LearningVideo GenerationConsistencyArtificial IntelligenceMultimodalGenerative AI
02

OmniGAIA: Towards Native Omni-Modal AI Agents

Human intelligence naturally intertwines omni-modal perception -- spanning vision, audio, and language -- with complex reasoning and tool usage to interact with the world. However, current multi-modal LLMs are primarily confined to bi-modal interactions (e.g., vision-language), lacking the unified cognitive capabilities required for general AI assistants. To bridge this gap, we introduce OmniGAIA, a comprehensive benchmark designed to evaluate omni-modal agents on tasks necessitating deep reasoning and multi-turn tool execution across video, audio, and image modalities. Constructed via a novel omni-modal event graph approach, OmniGAIA synthesizes complex, multi-hop queries derived from real-world data that require cross-modal reasoning and external tool integration. Furthermore, we propose OmniAtlas, a native omni-modal foundation agent under tool-integrated reasoning paradigm with active omni-modal perception. Trained on trajectories synthesized via a hindsight-guided tree exploration strategy and OmniDPO for fine-grained error correction, OmniAtlas effectively enhances the tool-use capabilities of existing open-source models. This work marks a step towards next-generation native omni-modal AI assistants for real-world scenarios.

Omni-Modal AI AgentsMulti-modal LLMsTool IntegrationCross-Modal ReasoningAI BenchmarksAI AgentMultimodalLarge Language Model
03

Risk-Aware World Model Predictive Control for Generalizable End-to-End Autonomous Driving

With advances in imitation learning (IL) and large-scale driving datasets, end-to-end autonomous driving (E2E-AD) has made great progress recently. Currently, IL-based methods have become a mainstream paradigm: models rely on standard driving behaviors given by experts, and learn to minimize the discrepancy between their actions and expert actions. However, this objective of "only driving like the expert" suffers from limited generalization: when encountering rare or unseen long-tail scenarios outside the distribution of expert demonstrations, models tend to produce unsafe decisions in the absence of prior experience. This raises a fundamental question: Can an E2E-AD system make reliable decisions without any expert action supervision? Motivated by this, we propose a unified framework named Risk-aware World Model Predictive Control (RaWMPC) to address this generalization dilemma through robust control, without reliance on expert demonstrations. Practically, RaWMPC leverages a world model to predict the consequences of multiple candidate actions and selects low-risk actions through explicit risk evaluation. To endow the world model with the ability to predict the outcomes of risky driving behaviors, we design a risk-aware interaction strategy that systematically exposes the world model to hazardous behaviors, making catastrophic outcomes predictable and thus avoidable. Furthermore, to generate low-risk candidate actions at test time, we introduce a self-evaluation distillation method to distill riskavoidance capabilities from the well-trained world model into a generative action proposal network without any expert demonstration. Extensive experiments show that RaWMPC outperforms state-of-the-art methods in both in-distribution and out-of-distribution scenarios, while providing superior decision interpretability.

Autonomous DrivingWorld ModelModel Predictive ControlRisk-AwareGeneralizationArtificial IntelligenceMachine LearningRobotics
04

Accelerating Diffusion via Hybrid Data-Pipeline Parallelism Based on Conditional Guidance Scheduling

Diffusion models have achieved remarkable progress in high-fidelity image, video, and audio generation, yet inference remains computationally expensive. Nevertheless, current diffusion acceleration methods based on distributed parallelism suffer from noticeable generation artifacts and fail to achieve substantial acceleration proportional to the number of GPUs. Therefore, we propose a hybrid parallelism framework that combines a novel data parallel strategy, condition-based partitioning, with an optimal pipeline scheduling method, adaptive parallelism switching, to reduce generation latency and achieve high generation quality in conditional diffusion models. The key ideas are to (i) leverage the conditional and unconditional denoising paths as a new data-partitioning perspective and (ii) adaptively enable optimal pipeline parallelism according to the denoising discrepancy between these two paths. Our framework achieves 2.31times and 2.07times latency reductions on SDXL and SD3, respectively, using two NVIDIA RTX~3090 GPUs, while preserving image quality. This result confirms the generality of our approach across U-Net-based diffusion models and DiT-based flow-matching architectures. Our approach also outperforms existing methods in acceleration under high-resolution synthesis settings. Code is available at https://github.com/kaist-dmlab/Hybridiff.

Diffusion modelsHybrid parallelismConditional guidance schedulingLatency reductionImage generationDeep LearningComputer VisionGenerative AI
05

Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning

Reinforcement Learning with Verifiable Rewards (RLVR) has become the leading paradigm for enhancing reasoning in Large Language Models (LLMs). However, standard RLVR algorithms suffer from a well-documented pathology: while they improve Pass@1 accuracy through sharpened sampling, they simultaneously narrow the model's reasoning boundary and reduce generation diversity. We identify a root cause that existing methods overlook: the uniform penalization of errors. Current approaches -- whether data-filtering methods that select prompts by difficulty, or advantage normalization schemes -- treat all incorrect rollouts within a group identically. We show that this uniformity allows overconfident errors (incorrect reasoning paths that the RL process has spuriously reinforced) to persist and monopolize probability mass, ultimately suppressing valid exploratory trajectories. To address this, we propose the Asymmetric Confidence-aware Error Penalty (ACE). ACE introduces a per-rollout confidence shift metric, c_i = log(pi_theta(y_i|x) / pi_ref(y_i|x)), to dynamically modulate negative advantages. Theoretically, we demonstrate that ACE's gradient can be decomposed into the gradient of a selective regularizer restricted to overconfident errors, plus a well-characterized residual that partially moderates the regularizer's strength. We conduct extensive experiments fine-tuning Qwen2.5-Math-7B, Qwen3-8B-Base, and Llama-3.1-8B-Instruct on the DAPO-Math-17K dataset using GRPO and DAPO within the VERL framework. Evaluated on MATH-500 and AIME 2025, ACE composes seamlessly with existing methods and consistently improves the full Pass@k spectrum across all three model families and benchmarks.

Reinforcement LearningLarge Language ModelsAsymmetric Confidence PenaltiesError CorrectionReasoning EnhancementLarge Language ModelMachine LearningDeep Learning
06

No One Size Fits All: QueryBandits for Hallucination Mitigation

Advanced reasoning capabilities in Large Language Models (LLMs) have led to more frequent hallucinations; yet most mitigation work focuses on open-source models for post-hoc detection and parameter editing. The dearth of studies focusing on hallucinations in closed-source models is especially concerning, as they constitute the vast majority of models in institutional deployments. We introduce QueryBandits, a model-agnostic contextual bandit framework that adaptively learns online to select the optimal query-rewrite strategy by leveraging an empirically validated and calibrated reward function. Across 16 QA scenarios, our top QueryBandit (Thompson Sampling) achieves an 87.5% win rate over a No-Rewrite baseline and outperforms zero-shot static policies (e.g., Paraphrase or Expand) by 42.6% and 60.3%, respectively. Moreover, all contextual bandits outperform vanilla bandits across all datasets, with higher feature variance coinciding with greater variance in arm selection. This substantiates our finding that there is no single rewrite policy optimal for all queries. We also discover that certain static policies incur higher cumulative regret than No-Rewrite, indicating that an inflexible query-rewriting policy can worsen hallucinations. Thus, learning an online policy over semantic features with QueryBandits can shift model behavior purely through forward-pass mechanisms, enabling its use with closed-source models and bypassing the need for retraining or gradient-based adaptation.

Hallucination MitigationLarge Language ModelsQueryBanditsContextual BanditQuery RewritingLarge Language ModelMachine LearningNatural Language Processing