NO/FOMO

Independent AI signal, once a day

The AI briefing worth opening.

ISSUE DATE2025-10-31ENGLISH EDITION
This issue
—
All time
—

Hacker News

6 stories
01

Kimi Linear: An Expressive, Efficient Attention Architecture

Kimi Linear presents a novel attention architecture developed by MoonshotAI, specifically engineered to enhance both the expressiveness and efficiency of deep learning models. This innovation directly addresses the significant computational overhead associated with conventional attention mechanisms, which typically exhibit quadratic scaling with respect to input sequence length. By adopting a linear scaling approach, Kimi Linear aims to dramatically reduce the computational and memory requirements, making it particularly advantageous for processing very long sequences in advanced AI applications, such as large language models. The architecture is designed to capture complex contextual relationships effectively while significantly improving training and inference speed. This development is crucial for advancing the scalability and practical deployment of high-performance neural networks, offering a more resource-efficient pathway for developing increasingly sophisticated AI systems and fostering broader accessibility to powerful AI capabilities.

Attention MechanismDeep LearningLarge Language ModelsNeural NetworksComputational EfficiencyArtificial IntelligenceDeep LearningLarge Language Model
02

Reasoning models reason well, until they don't

This article discusses the inherent limitations and potential failure points of current AI reasoning models. While these models have demonstrated remarkable abilities in various complex tasks, their performance is not consistently robust across all scenarios, revealing critical vulnerabilities. The research likely explores the conditions under which these advanced models cease to reason effectively, highlighting specific edge cases, biases, or contextual shifts that lead to erroneous or illogical outputs. Understanding these boundaries is crucial for developing more reliable and trustworthy artificial intelligence systems. The findings underscore the need for further research into improving model robustness, enhancing their ability to handle ambiguous information, and ensuring more consistent and predictable reasoning capabilities in real-world applications.

AI ReasoningModel RobustnessFailure ModesArtificial IntelligenceMachine LearningArtificial IntelligenceMachine LearningDeep Learning
03

Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop

A recent Hacker News inquiry seeks practical insights into the local deployment and usage of open-source Large Language Models (LLMs) and coding assistants on personal laptops. The post invites developers and enthusiasts to share their real-world workflows, focusing on specific models (e.g., Ollama, LM Studio) and integration tools, particularly VS Code plugins. Key information requested includes detailed laptop hardware configurations—such as CPU, GPU/NPU type, and memory—along with performance assessments for these local AI setups. Furthermore, the discussion aims to uncover the diverse applications of these tools, encompassing tasks like code completion, refactoring, debugging, and code review, while also evaluating their reliability and identifying areas where they currently fall short. This investigation by Andrea aims to gather comprehensive data on the viability and efficacy of running generative AI tools directly on developer hardware, moving beyond cloud-based solutions.

Open-source LLMsCoding AssistantsLocal AI DeploymentDeveloper WorkflowsGenerative AILarge Language ModelGenerative AIArtificial Intelligence
04

How OpenAI uses complex and circular deals to fuel its multibillion-dollar rise

An investigative report explores the intricate financial architecture that underpins OpenAI's rapid expansion and its journey to a multibillion-dollar valuation. The article details how the prominent artificial intelligence company employs a sophisticated web of complex and often circular financial deals to continuously secure significant capital. These arrangements, which may involve interconnected investments, strategic partnerships, and unique funding structures, are presented as critical mechanisms enabling OpenAI to sustain its capital-intensive research and development in areas like large language models and generative AI. The analysis aims to shed light on the economic engineering behind OpenAI's market dominance, offering insights into its distinct business model and the innovative, albeit potentially opaque, financial strategies that propel its accelerated growth within the highly competitive AI industry.

OpenAIAI FundingVenture CapitalFinancial StrategyTech Business ModelArtificial IntelligenceLarge Language ModelGenerative AI
05

AI scrapers request commented scripts

Recent observations suggest that AI-powered web scrapers are showing a preference for, or even implicitly 'requesting,' commented scripts. This trend indicates that AI systems designed for data extraction and learning from web content are more effective when code is well-documented. Commented scripts likely provide richer contextual information, improving the AI's ability to parse, understand, and categorize code functionalities beyond mere syntax. This phenomenon could have significant implications for web developers and content creators, encouraging the practice of thorough code commenting not just for human readability but also for optimal machine comprehension and data quality for AI models. It highlights a potential shift in the value placed on documentation, as AI systems increasingly become primary consumers of online data, influencing how information is structured and presented on the web.

AI scrapersweb scrapingcommented scriptscode understandingAI data collectionArtificial IntelligenceMachine LearningNatural Language Processing
06

Claude outage

Anthropic's Claude AI service experienced an unexpected outage, leading to significant disruption for users attempting to access the large language model. The incident, promptly reported on Claude's official status page, indicated a period of unavailability or degraded performance across its various platforms. This event likely impacted a broad spectrum of applications and services reliant on Claude's advanced AI capabilities, including conversational agents, content generation tools, and developer integrations. While the initial report from Hacker News provided minimal content, the nature of an 'outage' implies a critical service interruption, typically stemming from underlying infrastructure issues, software bugs, or unexpected surges in demand. The status page (status.claude.com) serves as the authoritative source for real-time updates regarding monitoring, identification, and resolution efforts. Such incidents underscore the importance of robust reliability measures and transparent communication in critical AI service provision, as developers and businesses depend on consistent access to maintain their operations and user experiences.

AI OutageService DisruptionLarge Language ModelsSystem UptimeCloud InfrastructureArtificial IntelligenceLarge Language ModelNatural Language Processing

GitHub

1 story
01

Claude Relay Service

The Claude Relay Service provides a self-hosted API relay solution for Claude, Codex, and Gemini, addressing common challenges such as regional restrictions, privacy concerns, cost-sharing among multiple users, and service instability. It enables users to deploy their own intermediary service, connecting directly to Anthropic's API while offering robust features like multi-account management with automatic rotation, custom API key generation for individual users, and detailed usage statistics. Advanced functionalities include intelligent account switching, performance optimization through connection pooling and caching, a comprehensive web-based monitoring dashboard, and granular security controls (access restrictions, rate limits, client-specific usage). The service also supports various proxy configurations (HTTP/SOCKS5) and flexible deployment options, including script-based, manual, and Docker-compose methods. This project is ideal for technical users seeking a stable, private, and cost-effective way to access leading LLM services, ensuring data security and transparent cost tracking.

Claude APIAI ProxyMulti-account ManagementAPI GatewayAnthropicArtificial IntelligenceLarge Language ModelAI Agent

huggingface

6 stories
01

The End of Manual Decoding: Towards Truly End-to-End Language Models

The "end-to-end" label for LLMs is a misnomer. In practice, they depend on a non-differentiable decoding process that requires laborious, hand-tuning of hyperparameters like temperature and top-p. This paper introduces AutoDeco, a novel architecture that enables truly "end-to-end" generation by learning to control its own decoding strategy. We augment the standard transformer with lightweight heads that, at each step, dynamically predict context-specific temperature and top-p values alongside the next-token logits. This approach transforms decoding into a parametric, token-level process, allowing the model to self-regulate its sampling strategy within a single forward pass. Through extensive experiments on eight benchmarks, we demonstrate that AutoDeco not only significantly outperforms default decoding strategies but also achieves performance comparable to an oracle-tuned baseline derived from "hacking the test set"-a practical upper bound for any static method. Crucially, we uncover an emergent capability for instruction-based decoding control: the model learns to interpret natural language commands (e.g., "generate with low randomness") and adjusts its predicted temperature and top-p on a token-by-token basis, opening a new paradigm for steerable and interactive LLM decoding.

End-to-End Language ModelsDecoding StrategyAutoDecoDynamic HyperparametersInstruction-Based ControlLarge Language ModelNatural Language ProcessingDeep Learning
02

Emu3.5: Native Multimodal Models are World Learners

We introduce Emu3.5, a large-scale multimodal world model that natively predicts the next state across vision and language. Emu3.5 is pre-trained end-to-end with a unified next-token prediction objective on a corpus of vision-language interleaved data containing over 10 trillion tokens, primarily derived from sequential frames and transcripts of internet videos. The model naturally accepts interleaved vision-language inputs and generates interleaved vision-language outputs. Emu3.5 is further post-trained with large-scale reinforcement learning to enhance multimodal reasoning and generation. To improve inference efficiency, we propose Discrete Diffusion Adaptation (DiDA), which converts token-by-token decoding into bidirectional parallel prediction, accelerating per-image inference by about 20x without sacrificing performance. Emu3.5 exhibits strong native multimodal capabilities, including long-horizon vision-language generation, any-to-image (X2I) generation, and complex text-rich image generation. It also exhibits generalizable world-modeling abilities, enabling spatiotemporally consistent world exploration and open-world embodied manipulation across diverse scenarios and tasks. For comparison, Emu3.5 achieves performance comparable to Gemini 2.5 Flash Image (Nano Banana) on image generation and editing tasks and demonstrates superior results on a suite of interleaved generation tasks. We open-source Emu3.5 at https://github.com/baaivision/Emu3.5 to support community research.

multimodal modelsworld modelvision-language generationreinforcement learningDiscrete Diffusion AdaptationMultimodalGenerative AIDeep Learning
03

Kimi Linear: An Expressive, Efficient Attention Architecture

We introduce Kimi Linear, a hybrid linear attention architecture that, for the first time, outperforms full attention under fair comparisons across various scenarios -- including short-context, long-context, and reinforcement learning (RL) scaling regimes. At its core lies Kimi Delta Attention (KDA), an expressive linear attention module that extends Gated DeltaNet with a finer-grained gating mechanism, enabling more effective use of limited finite-state RNN memory. Our bespoke chunkwise algorithm achieves high hardware efficiency through a specialized variant of the Diagonal-Plus-Low-Rank (DPLR) transition matrices, which substantially reduces computation compared to the general DPLR formulation while remaining more consistent with the classical delta rule. We pretrain a Kimi Linear model with 3B activated parameters and 48B total parameters, based on a layerwise hybrid of KDA and Multi-Head Latent Attention (MLA). Our experiments show that with an identical training recipe, Kimi Linear outperforms full MLA with a sizeable margin across all evaluated tasks, while reducing KV cache usage by up to 75% and achieving up to 6 times decoding throughput for a 1M context. These results demonstrate that Kimi Linear can be a drop-in replacement for full attention architectures with superior performance and efficiency, including tasks with longer input and output lengths. To support further research, we open-source the KDA kernel and vLLM implementations, and release the pre-trained and instruction-tuned model checkpoints.

Kimi LinearLinear AttentionAttention ArchitectureKimi Delta AttentionTransformer EfficiencyDeep LearningNatural Language ProcessingLarge Language Model
04

The Era of Agentic Organization: Learning to Organize with Language Models

We envision a new era of AI, termed agentic organization, where agents solve complex problems by working collaboratively and concurrently, enabling outcomes beyond individual intelligence. To realize this vision, we introduce asynchronous thinking (AsyncThink) as a new paradigm of reasoning with large language models, which organizes the internal thinking process into concurrently executable structures. Specifically, we propose a thinking protocol where an organizer dynamically assigns sub-queries to workers, merges intermediate knowledge, and produces coherent solutions. More importantly, the thinking structure in this protocol can be further optimized through reinforcement learning. Experiments demonstrate that AsyncThink achieves 28% lower inference latency compared to parallel thinking while improving accuracy on mathematical reasoning. Moreover, AsyncThink generalizes its learned asynchronous thinking capabilities, effectively tackling unseen tasks without additional training.

Agentic OrganizationAsynchronous ThinkingLarge Language ModelsReinforcement LearningAI AgentsArtificial IntelligenceLarge Language ModelAI Agent
05

The Quest for Generalizable Motion Generation: Data, Model, and Evaluation

Despite recent advances in 3D human motion generation (MoGen) on standard benchmarks, existing models still face a fundamental bottleneck in their generalization capability. In contrast, adjacent generative fields, most notably video generation (ViGen), have demonstrated remarkable generalization in modeling human behaviors, highlighting transferable insights that MoGen can leverage. Motivated by this observation, we present a comprehensive framework that systematically transfers knowledge from ViGen to MoGen across three key pillars: data, modeling, and evaluation. First, we introduce ViMoGen-228K, a large-scale dataset comprising 228,000 high-quality motion samples that integrates high-fidelity optical MoCap data with semantically annotated motions from web videos and synthesized samples generated by state-of-the-art ViGen models. The dataset includes both text-motion pairs and text-video-motion triplets, substantially expanding semantic diversity. Second, we propose ViMoGen, a flow-matching-based diffusion transformer that unifies priors from MoCap data and ViGen models through gated multimodal conditioning. To enhance efficiency, we further develop ViMoGen-light, a distilled variant that eliminates video generation dependencies while preserving strong generalization. Finally, we present MBench, a hierarchical benchmark designed for fine-grained evaluation across motion quality, prompt fidelity, and generalization ability. Extensive experiments show that our framework significantly outperforms existing approaches in both automatic and human evaluations. The code, data, and benchmark will be made publicly available.

Motion GenerationVideo GenerationGeneralizationDiffusion ModelsDatasetsGenerative AIDeep LearningMultimodal
06

Surfer 2: The Next Generation of Cross-Platform Computer Use Agents

Building agents that generalize across web, desktop, and mobile environments remains an open challenge, as prior systems rely on environment-specific interfaces that limit cross-platform deployment. We introduce Surfer 2, a unified architecture operating purely from visual observations that achieves state-of-the-art performance across all three environments. Surfer 2 integrates hierarchical context management, decoupled planning and execution, and self-verification with adaptive recovery, enabling reliable operation over long task horizons. Our system achieves 97.1% accuracy on WebVoyager, 69.6% on WebArena, 60.1% on OSWorld, and 87.1% on AndroidWorld, outperforming all prior systems without task-specific fine-tuning. With multiple attempts, Surfer 2 exceeds human performance on all benchmarks. These results demonstrate that systematic orchestration amplifies foundation model capabilities and enables general-purpose computer control through visual interaction alone, while calling for a next-generation vision language model to achieve Pareto-optimal cost-efficiency.

Cross-Platform AgentsVisual ObservationsFoundation ModelsComputer ControlAI AgentAI AgentComputer VisionMultimodal