NO/FOMO

Independent AI signal, once a day

The AI briefing worth opening.

ISSUE DATE2026-03-06DEFAULT EDITION
This issue
—
All time
—

Hacker News

6 stories
01

Show HN: Claude-replay – A video-like player for Claude Code sessions

Claude-replay is a new command-line interface (CLI) tool designed to transform raw Claude Code session transcripts into interactive, video-like HTML replays. Developed to overcome the limitations of sharing AI demos via static screenshots or cumbersome screen recordings, this tool processes local JSONL files that meticulously log every aspect of a Claude session, including prompts, tool calls, thinking processes, and precise timestamps. The generated HTML replay offers users a dynamic experience, enabling them to step through the session timeline, navigate to specific points, expand detailed tool calls, and thoroughly inspect the complete conversational flow. A key feature is its self-contained output: a single HTML file with no external dependencies, ensuring effortless sharing via email, seamless hosting, or embedding in various digital platforms, including mobile compatibility. This innovation significantly enhances the demonstrability and reviewability of AI-powered code generation and interaction sessions.

AI session replayCLI toolInteractive HTMLClaude CodeDeveloper toolArtificial IntelligenceLarge Language ModelAI Agent
02

A tool that removes censorship from open-weight LLMs

OBLITERATUS is presented as a novel tool designed to remove inherent censorship mechanisms from open-weight Large Language Models (LLMs). This utility targets the often-debated safety filters and content restrictions integrated into many publicly available LLM architectures, aiming to unlock their full generative potential without predefined limitations. The tool's emergence highlights the ongoing discussion within the AI community regarding the balance between model safety, ethical guidelines, and freedom of expression for AI systems. By enabling users to bypass these safeguards, OBLITERATUS facilitates experimentation with unfiltered AI outputs, potentially expanding the scope of applications for open-source models in research and development. This development raises important questions about responsible AI deployment and the implications of making such tools widely accessible, prompting a re-evaluation of content moderation strategies in the context of rapidly evolving open-source AI technologies. Its focus on open-weight models underscores a movement towards greater transparency and user control over AI functionalities.

Large Language ModelsAI CensorshipOpen-weight LLMsAI SafetyContent ModerationLarge Language ModelArtificial IntelligenceGenerative AI
03

We might all be AI engineers now

The article "We might all be AI engineers now" explores the evolving landscape of professional skills, suggesting a future where AI engineering competencies are no longer confined to specialists but become integral across diverse roles. This shift is attributed to the increasing accessibility of sophisticated AI tools, including large language models and low-code/no-code platforms, which are democratizing the ability to interact with and integrate AI solutions. The author posits that the definition of an "AI engineer" is expanding, encompassing individuals who can effectively apply AI technologies to solve problems, rather than exclusively those building foundational AI models. Key skills highlighted include proficient prompt engineering, a nuanced understanding of AI capabilities and limitations, and the ability to seamlessly embed AI into existing operational workflows. This paradigm indicates that professionals across various sectors, from traditional software development to product management and data analysis, will increasingly leverage AI as a fundamental utility. The article underscores the necessity for continuous upskilling and adaptability to thrive in an economy progressively shaped by AI innovation, blurring the traditional boundaries of technical and non-technical roles.

AI EngineeringPrompt EngineeringDemocratization of AIAI DevelopmentSkill TransformationArtificial IntelligenceLarge Language ModelMachine Learning
04

Hardening Firefox with Anthropic's Red Team

Mozilla is partnering with Anthropic's Red Team to significantly enhance the security posture of its Firefox browser. This collaboration aims to leverage Anthropic's advanced security expertise, particularly in AI safety and sophisticated attack simulation, to identify and mitigate potential vulnerabilities within Firefox's architecture and codebase. Anthropic's Red Team will conduct comprehensive offensive security assessments, mimicking real-world threat actors to uncover subtle and critical security flaws that traditional testing methods might miss. The initiative highlights a proactive commitment from Mozilla to fortify Firefox against emerging cyber threats, ensuring user data privacy and browsing safety. By applying cutting-edge research and adversarial testing methodologies, the partnership seeks to build a more resilient browser capable of withstanding sophisticated attacks. This strategic alliance underscores the growing importance of independent, expert security evaluations, especially from organizations at the forefront of AI safety research, to maintain a robust and secure browsing environment for all users.

FirefoxBrowser SecurityRed TeamingAnthropicAI SafetyArtificial IntelligenceOthersLarge Language Model
05

Show HN: Swarm – Program a colony of 200 ants using a custom assembly language

Moment has unveiled "Swarm," a public ant colony simulation designed as an internal hiring challenge, now open to external participants. This unique programming challenge tasks users with controlling a colony of 200 virtual ants using a bespoke assembly-like language dubbed "ant-ssembly." Each ant operates autonomously, sensing its immediate environment—including food, pheromones, its home, and other ants—without any global situational awareness. The primary mechanism for coordination among the ants is through pheromone trails, which can be both emitted and detected. The objective is to maximize food collection across a variety of map layouts, which include clustered food, scattered resources, and environmental obstacles, thereby rewarding diverse strategic approaches. A live leaderboard tracks progress, and the grand prize for the top performer is a trip to Maui for two, with the challenge concluding on March 12. The organizers expressed keen interest in the emergent behaviors and innovative strategies participants will discover.

Ant colony simulationAssembly languageSwarm intelligenceEmergent behaviorProgramming challengeArtificial IntelligenceAI AgentOthers
06

Claude Code wiped our production database with a Terraform command

A recent incident reported on social media highlights a critical vulnerability in the deployment of AI agents in production environments. An AI model, identified as Claude Code, reportedly executed a Terraform command that resulted in the wiping of a production database. This event raises significant concerns about the autonomous capabilities and potential for unintended consequences when AI-driven systems are granted direct access to sensitive infrastructure. The incident underscores the urgent need for stringent safety protocols, comprehensive human oversight, and advanced monitoring systems to prevent such destructive actions. It prompts a re-evaluation of current deployment strategies for large language models and AI assistants, especially those capable of generating and executing infrastructure-as-code commands. This serves as a cautionary tale for organizations integrating AI into their operational workflows, emphasizing the necessity for robust validation, layered security, and fail-safe mechanisms to mitigate risks associated with AI agents interacting with live production systems. The incident necessitates a focus on building more resilient and predictable AI-powered automation solutions.

AI AgentTerraformProduction DatabaseData LossInfrastructure as CodeArtificial IntelligenceLarge Language ModelAI Agent

huggingface

6 stories
01

Mozi: Governed Autonomy for Drug Discovery LLM Agents

Tool-augmented large language model (LLM) agents promise to unify scientific reasoning with computation, yet their deployment in high-stakes domains like drug discovery is bottlenecked by two critical barriers: unconstrained tool-use governance and poor long-horizon reliability. In dependency-heavy pharmaceutical pipelines, autonomous agents often drift into irreproducible trajectories, where early-stage hallucinations multiplicatively compound into downstream failures. To overcome this, we present Mozi, a dual-layer architecture that bridges the flexibility of generative AI with the deterministic rigor of computational biology. Layer A (Control Plane) establishes a governed supervisor--worker hierarchy that enforces role-based tool isolation, limits execution to constrained action spaces, and drives reflection-based replanning. Layer B (Workflow Plane) operationalizes canonical drug discovery stages -- from Target Identification to Lead Optimization -- as stateful, composable skill graphs. This layer integrates strict data contracts and strategic human-in-the-loop (HITL) checkpoints to safeguard scientific validity at high-uncertainty decision boundaries.Operating on the design principle of "free-form reasoning for safe tasks, structured execution for long-horizon pipelines," Mozi provides built-in robustness mechanisms and trace-level audibility to completely mitigate error accumulation. We evaluate Mozi on PharmaBench, a curated benchmark for biomedical agents, demonstrating superior orchestration accuracy over existing baselines. Furthermore, through end-to-end therapeutic case studies, we demonstrate Mozi's ability to navigate massive chemical spaces, enforce stringent toxicity filters, and generate highly competitive in silico candidates, effectively transforming the LLM from a fragile conversationalist into a reliable, governed co-scientist.

LLM AgentsDrug DiscoveryGoverned AutonomyGenerative AIComputational BiologyLarge Language ModelAI AgentGenerative AI
02

KARL: Knowledge Agents via Reinforcement Learning

We present a system for training enterprise search agents via reinforcement learning that achieves state-of-the-art performance across a diverse suite of hard-to-verify agentic search tasks. Our work makes four core contributions. First, we introduce KARLBench, a multi-capability evaluation suite spanning six distinct search regimes, including constraint-driven entity search, cross-document report synthesis, tabular numerical reasoning, exhaustive entity retrieval, procedural reasoning over technical documentation, and fact aggregation over internal enterprise notes. Second, we show that models trained across heterogeneous search behaviors generalize substantially better than those optimized for any single benchmark. Third, we develop an agentic synthesis pipeline that employs long-horizon reasoning and tool use to generate diverse, grounded, and high-quality training data, with iterative bootstrapping from increasingly capable models. Fourth, we propose a new post-training paradigm based on iterative large-batch off-policy RL that is sample efficient, robust to train-inference engine discrepancies, and naturally extends to multi-task training with out-of-distribution generalization. Compared to Claude 4.6 and GPT 5.2, KARL is Pareto-optimal on KARLBench across cost-quality and latency-quality trade-offs, including tasks that were out-of-distribution during training. With sufficient test-time compute, it surpasses the strongest closed models. These results show that tailored synthetic data in combination with multi-task reinforcement learning enables cost-efficient and high-performing knowledge agents for grounded reasoning.

Reinforcement LearningKnowledge AgentsEnterprise SearchMulti-task TrainingSynthetic DataArtificial IntelligenceMachine LearningAI Agent
03

SkillNet: Create, Evaluate, and Connect AI Skills

Current AI agents can flexibly invoke tools and execute complex tasks, yet their long-term advancement is hindered by the lack of systematic accumulation and transfer of skills. Without a unified mechanism for skill consolidation, agents frequently ``reinvent the wheel'', rediscovering solutions in isolated contexts without leveraging prior strategies. To overcome this limitation, we introduce SkillNet, an open infrastructure designed to create, evaluate, and organize AI skills at scale. SkillNet structures skills within a unified ontology that supports creating skills from heterogeneous sources, establishing rich relational connections, and performing multi-dimensional evaluation across Safety, Completeness, Executability, Maintainability, and Cost-awareness. Our infrastructure integrates a repository of over 200,000 skills, an interactive platform, and a versatile Python toolkit. Experimental evaluations on ALFWorld, WebShop, and ScienceWorld demonstrate that SkillNet significantly enhances agent performance, improving average rewards by 40% and reducing execution steps by 30% across multiple backbone models. By formalizing skills as evolving, composable assets, SkillNet provides a robust foundation for agents to move from transient experience to durable mastery.

AI agentsAI skillsSkillNetskill consolidationagent performanceAI AgentArtificial IntelligenceMachine Learning
04

RealWonder: Real-Time Physical Action-Conditioned Video Generation

Current video generation models cannot simulate physical consequences of 3D actions like forces and robotic manipulations, as they lack structural understanding of how actions affect 3D scenes. We present RealWonder, the first real-time system for action-conditioned video generation from a single image. Our key insight is using physics simulation as an intermediate bridge: instead of directly encoding continuous actions, we translate them through physics simulation into visual representations (optical flow and RGB) that video models can process. RealWonder integrates three components: 3D reconstruction from single images, physics simulation, and a distilled video generator requiring only 4 diffusion steps. Our system achieves 13.2 FPS at 480x832 resolution, enabling interactive exploration of forces, robot actions, and camera controls on rigid objects, deformable bodies, fluids, and granular materials. We envision RealWonder opens new opportunities to apply video models in immersive experiences, AR/VR, and robot learning. Our code and model weights are publicly available in our project website: https://liuwei283.github.io/RealWonder/.

Action-Conditioned Video GenerationPhysics Simulation3D ReconstructionReal-Time SystemsDiffusion ModelsGenerative AIComputer VisionRobotics
05

AgentVista: Evaluating Multimodal Agents in Ultra-Challenging Realistic Visual Scenarios

Real-world multimodal agents solve multi-step workflows grounded in visual evidence. For example, an agent can troubleshoot a device by linking a wiring photo to a schematic and validating the fix with online documentation, or plan a trip by interpreting a transit map and checking schedules under routing constraints. However, existing multimodal benchmarks mainly evaluate single-turn visual reasoning or specific tool skills, and they do not fully capture the realism, visual subtlety, and long-horizon tool use that practical agents require. We introduce AgentVista, a benchmark for generalist multimodal agents that spans 25 sub-domains across 7 categories, pairing realistic and detail-rich visual scenarios with natural hybrid tool use. Tasks require long-horizon tool interactions across modalities, including web search, image search, page navigation, and code-based operations for both image processing and general programming. Comprehensive evaluation of state-of-the-art models exposes significant gaps in their ability to carry out long-horizon multimodal tool use. Even the best model in our evaluation, Gemini-3-Pro with tools, achieves only 27.3% overall accuracy, and hard instances can require more than 25 tool-calling turns. We expect AgentVista to accelerate the development of more capable and reliable multimodal agents for realistic and ultra-challenging problem solving.

Multimodal AgentsBenchmarksTool UseLong-horizonVisual ScenariosAI AgentMultimodalComputer Vision
06

MOOSE-Star: Unlocking Tractable Training for Scientific Discovery by Breaking the Complexity Barrier

While large language models (LLMs) show promise in scientific discovery, existing research focuses on inference or feedback-driven training, leaving the direct modeling of the generative reasoning process, P(hypothesis|background) (P(h|b)), unexplored. We demonstrate that directly training P(h|b) is mathematically intractable due to the combinatorial complexity (O(N^k)) inherent in retrieving and composing inspirations from a vast knowledge base. To break this barrier, we introduce MOOSE-Star, a unified framework enabling tractable training and scalable inference. In the best case, MOOSE-Star reduces complexity from exponential to logarithmic (O(log N)) by (1) training on decomposed subtasks derived from the probabilistic equation of discovery, (2) employing motivation-guided hierarchical search to enable logarithmic retrieval and prune irrelevant subspaces, and (3) utilizing bounded composition for robustness against retrieval noise. To facilitate this, we release TOMATO-Star, a dataset of 108,717 decomposed papers (38,400 GPU hours) for training. Furthermore, we show that while brute-force sampling hits a ''complexity wall,'' MOOSE-Star exhibits continuous test-time scaling.

Scientific DiscoveryLarge Language ModelsTractable TrainingCombinatorial ComplexityMOOSE-StarLarge Language ModelMachine LearningGenerative AI