NO/FOMO

Independent AI signal, once a day

The AI briefing worth opening.

ISSUE DATE2025-10-29ENGLISH EDITION
This issue
—
All time
—

Hacker News

6 stories
01

Composer: Building a fast frontier model with RL

The article delves into 'Composer,' an innovative framework dedicated to the construction of rapid, state-of-the-art AI models, with a central emphasis on the application of Reinforcement Learning (RL). Composer is presented as a significant step forward in automating and optimizing the model development lifecycle, addressing the critical need for both high performance and computational efficiency in modern AI systems. By integrating advanced RL algorithms, Composer aims to autonomously discover and fine-tune model architectures, optimize training procedures, and adapt to diverse computational environments. This methodology allows for the iterative improvement of models, enabling them to reach 'frontier' performance benchmarks while simultaneously ensuring operational speed and resource effectiveness. The core idea is to leverage the adaptive learning capabilities of RL to push beyond the limitations of manual model design, fostering the creation of robust, scalable, and exceptionally fast AI solutions suitable for a wide array of demanding applications. This approach signifies a paradigm shift towards more intelligent and self-improving systems for AI model generation.

Reinforcement LearningModel OptimizationAI Model BuildingFrontier ModelsMachine LearningArtificial IntelligenceMachine LearningAI Agent
02

The end of the rip-off economy: consumers use LLMs against information asymmetry

Large Language Models (LLMs) are poised to fundamentally reshape consumer markets by directly addressing the long-standing issue of information asymmetry, potentially signaling "the end of the rip-off economy." Historically, businesses have often profited from consumers' lack of comprehensive information regarding pricing, product quality, and service terms. However, LLMs now empower consumers to rapidly process vast amounts of data, analyze complex contracts, compare offers efficiently, and identify hidden costs or unfavorable clauses. This technological shift enables individuals to make more informed decisions, fostering greater transparency and competition across various sectors, from financial services and healthcare to retail and automotive. The increased access to and interpretation of information by consumers through AI tools is compelling industries to adapt to a more level playing field, where value and fairness become paramount competitive advantages. This paradigm shift suggests a future where markets are driven by well-informed consumer choices, reducing the prevalence of exploitative practices.

Large Language ModelInformation AsymmetryConsumer EmpowermentEconomic TransparencyAI ApplicationsArtificial IntelligenceLarge Language ModelNatural Language Processing
03

Tell HN: Twilio support replies with hallucinated features

A Hacker News post highlights a critical issue regarding AI integration in customer support, where a Twilio user encountered "hallucinated features" from the company's support team. While investigating a bug in a voice system, the user was directed by support to non-existent debugging information and event logs within the Twilio interface. The user explicitly noted that the support message, containing these erroneous details, appeared to be "mostly AI written." This incident underscores the current limitations of AI in delivering accurate and reliable information, particularly in complex technical support scenarios. Despite ongoing discussions about the imminent arrival of Artificial General Intelligence (AGI), this real-world example demonstrates that contemporary AI systems can still produce unreliable outputs, leading to user frustration and inefficiency. The post criticizes the disparity between ambitious AGI claims and the practical, often flawed, application of current AI technologies, emphasizing the need for robust verification and human oversight in AI-powered support channels.

AI HallucinationsCustomer Support AITwilioLarge Language ModelsAI LimitationsArtificial IntelligenceLarge Language ModelAI Agent
04

Cursor 2.0

Cursor, an AI-native code editor, has officially rolled out its 2.0 release, marking a significant milestone in AI-powered software development tools. This major update likely introduces a suite of advanced features aimed at enhancing developer productivity and streamlining the coding workflow. While specific details are expected to be elaborated in the associated changelog, the 2.0 version is anticipated to offer improved AI capabilities such as more sophisticated code generation, intelligent debugging assistance, and context-aware refactoring tools. The update may also focus on optimizing performance, broadening language support, and refining the user experience with new interface elements or deeper integration points. This release underscores Cursor's commitment to leveraging cutting-edge artificial intelligence to transform the traditional development environment into a more efficient and intelligent workspace, further solidifying its position in the competitive landscape of developer tools.

AI Code EditorIntegrated Development EnvironmentCode GenerationDeveloper ToolsAI ProgrammingArtificial IntelligenceAI AgentGenerative AI
05

Extropic is building thermodynamic computing hardware

Extropic is actively pursuing the development of thermodynamic computing hardware, representing a significant departure from traditional computational paradigms. This novel approach harnesses principles of statistical mechanics and thermodynamics to execute computations, aiming to address the escalating energy demands and performance ceilings encountered by both classical and quantum computing systems. By embracing the inherent probabilistic and noisy characteristics of physical processes, rather than trying to mitigate them, thermodynamic computing seeks to create highly energy-efficient and scalable hardware. Such architectures are particularly promising for complex computational challenges prevalent in artificial intelligence and machine learning, including optimization problems, probabilistic inference, and efficient sampling. This initiative underscores a broader industry trend towards exploring alternative physics-based computing methods to pave the way for future advancements in AI acceleration and beyond, potentially unlocking new frontiers in computational power and efficiency.

Thermodynamic ComputingComputing HardwareAI HardwareNovel ArchitecturesEnergy EfficiencyArtificial IntelligenceMachine LearningOthers
06

Building a Robot Dog (with an airsoft gun)

This Hacker News story outlines a significant engineering project focused on the development of a quadruped robot dog, uniquely enhanced with the integration of an airsoft gun. The endeavor delves into multiple critical technical domains, including advanced mechanical design to ensure robust locomotion and structural integrity, sophisticated electrical engineering for efficient power distribution and sensor network management, and intricate software development covering real-time control systems, perception, and potentially autonomous navigation algorithms. A central challenge revolves around achieving dynamic stability and precise operational control, particularly when managing the added weight and functional requirements of the airsoft mechanism. This initiative serves as a compelling demonstration of applied robotics, illustrating the complexities inherent in mechatronic system design, component integration, and the creation of functional, intelligent agents. The project exemplifies how advanced concepts in robotics can be realized through hands-on development, pushing the frontiers of experimental hardware and integrated control for dynamic, autonomous tasks.

RoboticsQuadruped RobotMechatronicsEmbedded SystemsHardware DevelopmentRoboticsArtificial IntelligenceAI Agent

GitHub

2 stories
01

Handy

Handy is a free, open-source, and extensible cross-platform desktop application designed for privacy-focused, offline speech-to-text transcription. Built with Tauri, combining Rust for backend logic and React/TypeScript for the frontend, Handy allows users to transcribe spoken words into any text field simply by pressing a configurable keyboard shortcut, all without sending audio to the cloud. Its core functionality involves voice activity detection (VAD) using Silero, followed by transcription via either Whisper models (supporting GPU acceleration for Small/Medium/Turbo/Large variants) or the CPU-optimized Parakeet V3 model, which also offers automatic language detection. The project emphasizes its "forkable" nature, aiming to provide a simple, well-patterned codebase for community contributions and further development. Handy supports Windows, macOS, and Linux, making advanced speech recognition accessible and private, especially useful for accessibility tooling and enhancing productivity across various hardware setups.

Speech-to-textOffline transcriptionTauriRustWhisperArtificial IntelligenceNatural Language ProcessingMachine Learning
02

Web Development for Beginners - A Curriculum

This Microsoft Cloud Advocates' 'Web Development for Beginners' is a comprehensive 12-week curriculum designed to teach the fundamentals of web development. Comprising 24 lessons, the course covers JavaScript, CSS, and HTML through hands-on projects, including building terrariums, browser extensions, and space games. The pedagogy emphasizes project-based learning, supplemented by quizzes, discussions, and practical assignments to enhance skill development and knowledge retention. The curriculum also integrates new challenges leveraging GitHub Copilot Agent mode and introduces a Generative AI project, along with a newly released Generative AI for JavaScript course. Learners can set up their environment using GitHub Codespaces or locally with Visual Studio Code. It provides extensive multi-language support and encourages community engagement through Discord and discussion forums.

Web DevelopmentJavaScriptHTMLCSSGenerative AIOthersGenerative AIArtificial Intelligence

huggingface

6 stories
01

Tongyi DeepResearch Technical Report

We present Tongyi DeepResearch, an agentic large language model, which is specifically designed for long-horizon, deep information-seeking research tasks. To incentivize autonomous deep research agency, Tongyi DeepResearch is developed through an end-to-end training framework that combines agentic mid-training and agentic post-training, enabling scalable reasoning and information seeking across complex tasks. We design a highly scalable data synthesis pipeline that is fully automatic, without relying on costly human annotation, and empowers all training stages. By constructing customized environments for each stage, our system enables stable and consistent interactions throughout. Tongyi DeepResearch, featuring 30.5 billion total parameters, with only 3.3 billion activated per token, achieves state-of-the-art performance across a range of agentic deep research benchmarks, including Humanity's Last Exam, BrowseComp, BrowseComp-ZH, WebWalkerQA, xbench-DeepSearch, FRAMES and xbench-DeepSearch-2510. We open-source the model, framework, and complete solutions to empower the community.

Tongyi DeepResearchAgentic large language modelInformation-seeking researchEnd-to-end trainingDeep research benchmarksLarge Language ModelAI AgentDeep Learning
02

OSWorld-MCP: Benchmarking MCP Tool Invocation In Computer-Use Agents

With advances in decision-making and reasoning capabilities, multimodal agents show strong potential in computer application scenarios. Past evaluations have mainly assessed GUI interaction skills, while tool invocation abilities, such as those enabled by the Model Context Protocol (MCP), have been largely overlooked. Comparing agents with integrated tool invocation to those evaluated only on GUI interaction is inherently unfair. We present OSWorld-MCP, the first comprehensive and fair benchmark for assessing computer-use agents' tool invocation, GUI operation, and decision-making abilities in a real-world environment. We design a novel automated code-generation pipeline to create tools and combine them with a curated selection from existing tools. Rigorous manual validation yields 158 high-quality tools (covering 7 common applications), each verified for correct functionality, practical applicability, and versatility. Extensive evaluations of state-of-the-art multimodal agents on OSWorld-MCP show that MCP tools generally improve task success rates (e.g., from 8.3% to 20.4% for OpenAI o3 at 15 steps, from 40.1% to 43.3% for Claude 4 Sonnet at 50 steps), underscoring the importance of assessing tool invocation capabilities. However, even the strongest models have relatively low tool invocation rates, Only 36.3%, indicating room for improvement and highlighting the benchmark's challenge. By explicitly measuring MCP tool usage skills, OSWorld-MCP deepens understanding of multimodal agents and sets a new standard for evaluating performance in complex, tool-assisted environments. Our code, environment, and data are publicly available at https://osworld-mcp.github.io.

Computer-Use AgentsMultimodal AgentsTool InvocationBenchmarkingMCP ToolsAI AgentMultimodalArtificial Intelligence
03

Uniform Discrete Diffusion with Metric Path for Video Generation

Continuous-space video generation has advanced rapidly, while discrete approaches lag behind due to error accumulation and long-context inconsistency. In this work, we revisit discrete generative modeling and present Uniform discRete diffuSion with metric pAth (URSA), a simple yet powerful framework that bridges the gap with continuous approaches for the scalable video generation. At its core, URSA formulates the video generation task as an iterative global refinement of discrete spatiotemporal tokens. It integrates two key designs: a Linearized Metric Path and a Resolution-dependent Timestep Shifting mechanism. These designs enable URSA to scale efficiently to high-resolution image synthesis and long-duration video generation, while requiring significantly fewer inference steps. Additionally, we introduce an asynchronous temporal fine-tuning strategy that unifies versatile tasks within a single model, including interpolation and image-to-video generation. Extensive experiments on challenging video and image generation benchmarks demonstrate that URSA consistently outperforms existing discrete methods and achieves performance comparable to state-of-the-art continuous diffusion methods. Code and models are available at https://github.com/baaivision/URSA

Video GenerationDiscrete DiffusionGenerative ModelingSpatiotemporal TokensImage SynthesisGenerative AIComputer VisionDeep Learning
04

Agent Data Protocol: Unifying Datasets for Diverse, Effective Fine-tuning of LLM Agents

Public research results on large-scale supervised finetuning of AI agents remain relatively rare, since the collection of agent training data presents unique challenges. In this work, we argue that the bottleneck is not a lack of underlying data sources, but that a large variety of data is fragmented across heterogeneous formats, tools, and interfaces. To this end, we introduce the agent data protocol (ADP), a light-weight representation language that serves as an "interlingua" between agent datasets in diverse formats and unified agent training pipelines downstream. The design of ADP is expressive enough to capture a large variety of tasks, including API/tool use, browsing, coding, software engineering, and general agentic workflows, while remaining simple to parse and train on without engineering at a per-dataset level. In experiments, we unified a broad collection of 13 existing agent training datasets into ADP format, and converted the standardized ADP data into training-ready formats for multiple agent frameworks. We performed SFT on these data, and demonstrated an average performance gain of ~20% over corresponding base models, and delivers state-of-the-art or near-SOTA performance on standard coding, browsing, tool use, and research benchmarks, without domain-specific tuning. All code and data are released publicly, in the hope that ADP could help lower the barrier to standardized, scalable, and reproducible agent training.

Agent Data ProtocolLLM AgentsSupervised Fine-tuningData UnificationAgent TrainingAI AgentLarge Language ModelDeep Learning
05

Critique-RL: Training Language Models for Critiquing through Two-Stage Reinforcement Learning

Training critiquing language models to assess and provide feedback on model outputs is a promising way to improve LLMs for complex reasoning tasks. However, existing approaches typically rely on stronger supervisors for annotating critique data. To address this, we propose Critique-RL, an online RL approach for developing critiquing language models without stronger supervision. Our approach operates on a two-player paradigm: the actor generates a response, the critic provides feedback, and the actor refines the response accordingly. We first reveal that relying solely on indirect reward signals from the actor's outputs for RL optimization often leads to unsatisfactory critics: while their helpfulness (i.e., providing constructive feedback) improves, the discriminability (i.e., determining whether a response is high-quality or not) remains poor, resulting in marginal performance gains. To overcome this, Critique-RL adopts a two-stage optimization strategy. In stage I, it reinforces the discriminability of the critic with direct rule-based reward signals; in stage II, it introduces indirect rewards based on actor refinement to improve the critic's helpfulness, while maintaining its discriminability via appropriate regularization. Extensive experiments across various tasks and models show that Critique-RL delivers substantial performance improvements. For example, it achieves a 9.02% gain on in-domain tasks and a 5.70% gain on out-of-domain tasks for Qwen2.5-7B, highlighting its potential.

Reinforcement LearningLanguage ModelsCritiquingLLMsTwo-Stage OptimizationLarge Language ModelNatural Language ProcessingDeep Learning
06

SAO-Instruct: Free-form Audio Editing using Natural Language

Generative models have made significant progress in synthesizing high-fidelity audio from short textual descriptions. However, editing existing audio using natural language has remained largely underexplored. Current approaches either require the complete description of the edited audio or are constrained to predefined edit instructions that lack flexibility. In this work, we introduce SAO-Instruct, a model based on Stable Audio Open capable of editing audio clips using any free-form natural language instruction. To train our model, we create a dataset of audio editing triplets (input audio, edit instruction, output audio) using Prompt-to-Prompt, DDPM inversion, and a manual editing pipeline. Although partially trained on synthetic data, our model generalizes well to real in-the-wild audio clips and unseen edit instructions. We demonstrate that SAO-Instruct achieves competitive performance on objective metrics and outperforms other audio editing approaches in a subjective listening study. To encourage future research, we release our code and model weights.

audio editingnatural language instructiongenerative modelsSAO-InstructStable Audio OpenGenerative AIMachine LearningNatural Language Processing