NO/FOMO

每天一次，过滤 AI 噪音

值得打开的
AI 日报。

发布日期2026-02-11中文版本
本期阅读
—
累计阅读
—

Hacker News

5 stories
01

Claude Code Is Being Dumbed Down

Recent discussions and anecdotal reports circulating within the developer community suggest a potential decline in the quality and sophistication of code generated by Claude, a prominent large language model developed by Anthropic. The sentiment, encapsulated by the phrase 'dumbed down,' indicates that the AI's programming output may be becoming simpler, less accurate, or less robust compared to previous iterations. This perceived degradation could have significant implications for developers and users who depend on Claude for generating complex algorithms, refactoring code, or assisting with intricate software development tasks. Potential reasons for such a trend might include strategic internal model updates, recalibrations to optimize computational resources and reduce inference costs, or a shift in fine-tuning objectives that inadvertently prioritizes broader accessibility or specific simpler use-cases over advanced coding capabilities. If these observations are substantiated, it raises important questions regarding the consistency, reliability, and long-term utility of AI-powered code generation tools, emphasizing the continuous challenge for AI providers to balance innovation, performance, cost-efficiency, and the integrity of their models amidst rapid technological evolution. This situation underscores the dynamic nature of AI model development and the ongoing need for transparency regarding model changes and their impact on user experience.

ClaudeCode GenerationLarge Language ModelAI PerformanceModel DegradationArtificial IntelligenceLarge Language ModelGenerative AI
02

GLM-5: From Vibe Coding to Agentic Engineering

The article "GLM-5: From Vibe Coding to Agentic Engineering" highlights a significant advancement in the application of large language models (LLMs) to software development, focusing on the introduction of GLM-5. This new iteration, presumably a sophisticated General Language Model, is presented as a catalyst for evolving development methodologies from informal "vibe coding" to a more structured and automated paradigm termed "Agentic Engineering." The core premise is that GLM-5 empowers AI agents to undertake increasingly complex and autonomous engineering tasks, thereby streamlining the entire software development lifecycle. This includes everything from design and coding to testing and deployment, driven by intelligent agents. The shift signifies a move towards leveraging advanced LLMs to build more reliable, efficient, and sophisticated AI systems, emphasizing an future where AI itself plays a central, agent-driven role in its own creation and refinement. This transition underscores the growing impact of powerful language models in transforming modern engineering practices.

Large Language ModelsAI AgentsSoftware EngineeringAI DevelopmentGenerative AIArtificial IntelligenceLarge Language ModelAI Agent
03

AI-First Company Memos

This resource, titled "AI-First Company Memos" and hosted on the "the-ai-native.company" platform, delves into the strategic framework and practical guidelines for organizations aiming to embed artificial intelligence at the core of their operations and culture. It outlines a vision where AI is not merely a tool but a fundamental driver of business strategy, product innovation, and operational efficiency. The concept revolves around designing systems, processes, and products with AI as the primary consideration, influencing everything from data infrastructure to talent acquisition. Such memos would articulate the imperative for leadership to champion AI integration, empower teams with AI-driven insights, and foster an organizational mindset that prioritizes AI literacy and experimentation. This proactive adoption strategy is presented as essential for achieving sustainable competitive advantage, enabling rapid decision-making, and unlocking new avenues for growth in a rapidly evolving, technology-driven market.

AI StrategyEnterprise AIDigital TransformationOrganizational ChangeAI AdoptionArtificial IntelligenceMachine LearningAI Agent
04

End of an era for me: no more self-hosted git

An individual has publicly announced the cessation of their self-hosted Git repository, framing this decision as "the end of an era" for their personal development workflow. The accompanying blog post, titled "thank-you-ai," strongly implies that artificial intelligence or AI-powered tools and managed services have played a significant role in this transition. This move signifies a broader industry shift where developers are increasingly migrating from traditional, self-managed version control systems to cloud-based or commercially hosted solutions. These modern platforms often integrate advanced AI capabilities for features like automated code review, intelligent code suggestions, vulnerability detection, and streamlined deployment pipelines, significantly reducing the operational overhead associated with infrastructure maintenance. The discontinuation of self-hosting, therefore, highlights a strategic embrace of the efficiencies, scalability, and innovative functionalities offered by AI-augmented development platforms. This pivot allows the individual to reallocate resources and focus from managing underlying infrastructure to concentrating on core development tasks, underscoring the transformative impact of AI on contemporary software engineering practices and infrastructure choices.

Self-hosted GitVersion ControlArtificial IntelligenceDeveloper WorkflowCloud ServicesArtificial IntelligenceAI AgentOthers
05

GLM5 Released on Z.ai Platform

Z.ai Platform has officially announced the release of GLM5, introducing a new iteration of their general language model technology. This launch represents a significant development in the realm of artificial intelligence, likely offering enhanced capabilities in natural language understanding, generation, and complex reasoning tasks, building upon previous generations. The availability of GLM5 on the Z.ai platform suggests a strategic move to provide developers and enterprises with accessible, high-performance AI tools, potentially streamlining the deployment and integration of advanced language-centric applications across diverse industries. This release underscores the ongoing efforts within the AI community to refine and disseminate powerful models, fostering innovation across various sectors from content creation and data analysis to customer service and intelligent automation. The platform's role in hosting GLM5 aims to lower barriers to entry for integrating state-of-the-art AI, propelling further advancements in the practical application and adoption of large language models globally.

Large Language ModelAI PlatformModel ReleaseNatural Language ProcessingGenerative AIArtificial IntelligenceLarge Language ModelNatural Language Processing

huggingface

6 stories
01

OPUS: Towards Efficient and Principled Data Selection in Large Language Model Pre-training in Every Iteration

As high-quality public text approaches exhaustion, a phenomenon known as the Data Wall, pre-training is shifting from more tokens to better tokens. However, existing methods either rely on heuristic static filters that ignore training dynamics, or use dynamic yet optimizer-agnostic criteria based on raw gradients. We propose OPUS (Optimizer-induced Projected Utility Selection), a dynamic data selection framework that defines utility in the optimizer-induced update space. OPUS scores candidates by projecting their effective updates, shaped by modern optimizers, onto a target direction derived from a stable, in-distribution proxy. To ensure scalability, we employ Ghost technique with CountSketch for computational efficiency, and Boltzmann sampling for data diversity, incurring only 4.7% additional compute overhead. OPUS achieves remarkable results across diverse corpora, quality tiers, optimizers, and model scales. In pre-training of GPT-2 Large/XL on FineWeb and FineWeb-Edu with 30B tokens, OPUS outperforms industrial-level baselines and even full 200B-token training. Moreover, when combined with industrial-level static filters, OPUS further improves pre-training efficiency, even with lower-quality data. Furthermore, in continued pre-training of Qwen3-8B-Base on SciencePedia, OPUS achieves superior performance using only 0.5B tokens compared to full training with 3B tokens, demonstrating significant data efficiency gains in specialized domains.

Data SelectionLarge Language ModelPre-trainingOptimizerData EfficiencyLarge Language ModelDeep LearningNatural Language Processing
02

UI-Venus-1.5 Technical Report

GUI agents have emerged as a powerful paradigm for automating interactions in digital environments, yet achieving both broad generality and consistently strong task performance remains challenging.In this report, we present UI-Venus-1.5, a unified, end-to-end GUI Agent designed for robust real-world applications.The proposed model family comprises two dense variants (2B and 8B) and one mixture-of-experts variant (30B-A3B) to meet various downstream application scenarios.Compared to our previous version, UI-Venus-1.5 introduces three key technical advances: (1) a comprehensive Mid-Training stage leveraging 10 billion tokens across 30+ datasets to establish foundational GUI semantics; (2) Online Reinforcement Learning with full-trajectory rollouts, aligning training objectives with long-horizon, dynamic navigation in large-scale environments; and (3) a single unified GUI Agent constructed via Model Merging, which synthesizes domain-specific models (grounding, web, and mobile) into one cohesive checkpoint. Extensive evaluations demonstrate that UI-Venus-1.5 establishes new state-of-the-art performance on benchmarks such as ScreenSpot-Pro (69.6%), VenusBench-GD (75.0%), and AndroidWorld (77.6%), significantly outperforming previous strong baselines. In addition, UI-Venus-1.5 demonstrates robust navigation capabilities across a variety of Chinese mobile apps, effectively executing user instructions in real-world scenarios. Code: https://github.com/inclusionAI/UI-Venus; Model: https://huggingface.co/collections/inclusionAI/ui-venus

GUI AgentReinforcement LearningModel MergingTask AutomationDigital EnvironmentsAI AgentDeep LearningMachine Learning
03

Chain of Mindset: Reasoning with Adaptive Cognitive Modes

Human problem-solving is never the repetition of a single mindset, by which we mean a distinct mode of cognitive processing. When tackling a specific task, we do not rely on a single mindset; instead, we integrate multiple mindsets within the single solution process. However, existing LLM reasoning methods fall into a common trap: they apply the same fixed mindset across all steps, overlooking that different stages of solving the same problem require fundamentally different mindsets. This single-minded assumption prevents models from reaching the next level of intelligence. To address this limitation, we propose Chain of Mindset (CoM), a training-free agentic framework that enables step-level adaptive mindset orchestration. CoM decomposes reasoning into four functionally heterogeneous mindsets: Spatial, Convergent, Divergent, and Algorithmic. A Meta-Agent dynamically selects the optimal mindset based on the evolving reasoning state, while a bidirectional Context Gate filters cross-module information flow to maintain effectiveness and efficiency. Experiments across six challenging benchmarks spanning mathematics, code generation, scientific QA, and spatial reasoning demonstrate that CoM achieves state-of-the-art performance, outperforming the strongest baseline by 4.96% and 4.72% in overall accuracy on Qwen3-VL-32B-Instruct and Gemini-2.0-Flash, while balancing reasoning efficiency. Our code is publicly available at https://github.com/QuantaAlpha/chain-of-mindset.

Chain of MindsetLLM ReasoningAdaptive Cognitive ModesAI AgentMindset OrchestrationLarge Language ModelAI AgentArtificial Intelligence
04

Dr. MAS: Stable Reinforcement Learning for Multi-Agent LLM Systems

Multi-agent LLM systems enable advanced reasoning and tool use via role specialization, yet reliable reinforcement learning (RL) post-training for such systems remains difficult. In this work, we theoretically pinpoint a key reason for training instability when extending group-based RL to multi-agent LLM systems. We show that under GRPO-style optimization, a global normalization baseline may deviate from diverse agents' reward distributions, which ultimately leads to gradient-norm instability. Based on this finding, we propose Dr. MAS, a simple and stable RL training recipe for multi-agent LLM systems. Dr. MAS uses an agent-wise remedy: normalizing advantages per agent using each agent's own reward statistics, which calibrates gradient scales and dramatically stabilizes training, both theoretically and empirically. Beyond the algorithm, Dr. MAS provides an end-to-end RL training framework for multi-agent LLM systems, supporting scalable orchestration, flexible per-agent LLM serving and optimization configs, and shared resource scheduling of LLM actor backends. We evaluate Dr. MAS on multi-agent math reasoning and multi-turn search benchmarks using Qwen2.5 and Qwen3 series models. Dr. MAS achieves clear gains over vanilla GRPO (e.g., +5.6% avg@16 and +4.6% pass@16 on math, and +15.2% avg@16 and +13.1% pass@16 on search) while largely eliminating gradient spikes. Moreover, it remains highly effective under heterogeneous agent-model assignments while improving efficiency.

Multi-Agent LLM SystemsReinforcement LearningTraining StabilityGradient NormalizationAI AgentLarge Language ModelAI AgentMachine Learning
05

VideoWorld 2: Learning Transferable Knowledge from Real-world Videos

Learning transferable knowledge from unlabeled video data and applying it in new environments is a fundamental capability of intelligent agents. This work presents VideoWorld 2, which extends VideoWorld and offers the first investigation into learning transferable knowledge directly from raw real-world videos. At its core, VideoWorld 2 introduces a dynamic-enhanced Latent Dynamics Model (dLDM) that decouples action dynamics from visual appearance: a pretrained video diffusion model handles visual appearance modeling, enabling the dLDM to learn latent codes that focus on compact and meaningful task-related dynamics. These latent codes are then modeled autoregressively to learn task policies and support long-horizon reasoning. We evaluate VideoWorld 2 on challenging real-world handcraft making tasks, where prior video generation and latent-dynamics models struggle to operate reliably. Remarkably, VideoWorld 2 achieves up to 70% improvement in task success rate and produces coherent long execution videos. In robotics, we show that VideoWorld 2 can acquire effective manipulation knowledge from the Open-X dataset, which substantially improves task performance on CALVIN. This study reveals the potential of learning transferable world knowledge directly from raw videos, with all code, data, and models to be open-sourced for further research.

VideoWorld 2Transferable KnowledgeLatent Dynamics ModelVideo Diffusion ModelRoboticsRoboticsVideo UnderstandingDeep Learning
06

Autoregressive Image Generation with Masked Bit Modeling

This paper challenges the dominance of continuous pipelines in visual generation. We systematically investigate the performance gap between discrete and continuous methods. Contrary to the belief that discrete tokenizers are intrinsically inferior, we demonstrate that the disparity arises primarily from the total number of bits allocated in the latent space (i.e., the compression ratio). We show that scaling up the codebook size effectively bridges this gap, allowing discrete tokenizers to match or surpass their continuous counterparts. However, existing discrete generation methods struggle to capitalize on this insight, suffering from performance degradation or prohibitive training costs with scaled codebook. To address this, we propose masked Bit AutoRegressive modeling (BAR), a scalable framework that supports arbitrary codebook sizes. By equipping an autoregressive transformer with a masked bit modeling head, BAR predicts discrete tokens through progressively generating their constituent bits. BAR achieves a new state-of-the-art gFID of 0.99 on ImageNet-256, outperforming leading methods across both continuous and discrete paradigms, while significantly reducing sampling costs and converging faster than prior continuous approaches. Project page is available at https://bar-gen.github.io/

Autoregressive Image GenerationMasked Bit ModelingDiscrete Latent SpaceGenerative ModelsCodebook ScalingGenerative AIDeep LearningComputer Vision