NO/FOMO

Independent AI signal, once a day

The AI briefing worth opening.

ISSUE DATE2026-03-30ENGLISH EDITION
This issue
—
All time
—

Hacker News

6 stories
01

Copilot edited an ad into my PR

A developer reported an incident where GitHub Copilot, an AI-powered code completion tool, inserted an advertisement directly into a Pull Request (PR). The ad, for an unrelated product, appeared unexpectedly within the suggested code, raising significant concerns about the integrity and behavior of AI assistants in professional development environments. This event highlights potential risks associated with the commercialization of AI tools and the opaque nature of their operational mechanisms. It prompts discussions around data privacy, ethical AI guidelines, and the extent to which AI models might be influenced to inject sponsored content or exhibit unintended behaviors. The incident underscores the need for robust verification processes and clear policies regarding the content generated by AI co-pilots, ensuring they remain objective, secure, and focused solely on assisting developers without compromising code quality or introducing external agendas. This occurrence serves as a cautionary tale for the increasing integration of generative AI into critical software development workflows.

GitHub CopilotAI code generationGenerative AIDeveloper ToolsSoftware IntegrityArtificial IntelligenceGenerative AILarge Language Model
02

Hamilton-Jacobi-Bellman Equation: Reinforcement Learning and Diffusion Models

This analysis delves into the intricate relationship between the Hamilton-Jacobi-Bellman (HJB) equation, a cornerstone of optimal control theory, and two prominent areas of modern artificial intelligence: Reinforcement Learning (RL) and Diffusion Models. The HJB equation offers a profound mathematical framework for understanding and solving continuous-time optimal control problems, directly informing the development of value function-based methods in RL. By exploring this connection, the article illuminates how fundamental principles of dynamic programming and optimality can be applied to design more efficient and theoretically sound reinforcement learning algorithms. Furthermore, it investigates the emerging parallels between diffusion models, which excel at generating complex data distributions, and stochastic optimal control. This perspective suggests that diffusion processes can be reinterpreted through the lens of control, potentially leading to new insights into their mechanics or facilitating the integration of control-theoretic guarantees into generative AI. The work effectively bridges classical mathematical foundations with cutting-edge AI research.

Hamilton-Jacobi-Bellman EquationReinforcement LearningDiffusion ModelsOptimal ControlDynamic ProgrammingArtificial IntelligenceMachine LearningGenerative AI
03

Do your own writing

The article "Do your own writing," stemming from a discussion about the implications of AI in content creation, strongly advocates for the continued importance of human-led writing processes. It argues that over-reliance on artificial intelligence tools for generating text can impede the development of critical thinking, analytical skills, and a distinctive individual voice. While acknowledging the potential for AI to enhance productivity and assist with preliminary drafts, the core message emphasizes that true intellectual growth and the cultivation of effective communication abilities are rooted in personal effort and active engagement with the writing task. The piece implicitly warns against the erosion of human creativity and the unique cognitive benefits derived from constructing coherent arguments and expressive narratives. By urging individuals to "do your own writing," the author underscores the intrinsic value of developing and refining personal writing skills, viewing them as crucial for fostering deeper understanding, intellectual independence, and authentic expression in an increasingly automated world. This perspective serves as a counterpoint to the growing trend of delegating complex cognitive tasks to AI, highlighting the enduring necessity of human craftsmanship in communication.

Writing SkillsHuman CreativityAI EthicsCognitive DevelopmentCritical ThinkingArtificial IntelligenceNatural Language ProcessingLarge Language Model
04

The ladder is missing rungs – Engineering Progression When AI Ate the Middle

The article, "The ladder is missing rungs – Engineering Progression When AI Ate the Middle," explores the transformative impact of artificial intelligence on traditional engineering career paths. It posits that the increasing capabilities of AI and automation tools are significantly altering the demand for mid-level engineering roles, effectively "eating the middle" of the workforce. This phenomenon creates a critical challenge for professional development, as the conventional rungs of the career ladder, which previously offered clear progression from junior to senior positions through a series of intermediate tasks, are disappearing or becoming less defined. The author likely discusses how AI is automating routine and predictable engineering tasks, thereby reducing the opportunities for less experienced engineers to gain crucial experience and advance. Consequently, organizations face the imperative to rethink engineering progression models, emphasizing the need for upskilling in advanced areas like AI development, strategic problem-solving, and complex system design. The piece presumably advocates for a shift towards continuous learning and adaptation, urging engineers to acquire specialized skills to navigate this evolving landscape and secure a place in a workforce increasingly augmented by artificial intelligence.

AI in EngineeringWorkforce AutomationCareer ProgressionSkills GapFuture of WorkArtificial IntelligenceMachine LearningOthers
05

Show HN: Coasts – Containerized Hosts for Agents

Coasts, short for "containerized hosts," is a new tool designed to streamline development workflows by enabling the concurrent operation of multiple localhost instances and Docker Compose runtimes across various Git worktrees on a single machine. This addresses a significant challenge for developers, particularly those working with agents, who often need isolated testing environments for code changes made in different worktrees. Traditional methods involving port manipulation become unfeasible with complex Docker Compose configurations that feature numerous services and volumes. Coasts offers a solution by providing distinct, isolated, and worktree-scoped localhost runtimes, thereby simplifying the testing and development process for agents and complex applications. Demonstrations and conceptual overviews are available through a YouTube video and project documentation.

ContainerizationDocker ComposeGit WorktreesLocalhostDevelopment EnvironmentsAI AgentArtificial IntelligenceOthers
06

Mathematical methods and human thought in the age of AI

This article explores the intricate relationship between formal mathematical methods and the nuances of human thought within the rapidly evolving landscape of artificial intelligence. It delves into how mathematical foundations underpin the development and operation of advanced AI systems, from algorithms to neural network architectures, providing the rigor necessary for their functionality and reliability. Concurrently, the discussion examines the distinctive aspects of human cognition, including creativity, intuition, and ethical reasoning, which continue to pose unique challenges and opportunities in the AI era. The paper considers the potential for AI to augment human intellectual capabilities by processing vast datasets and identifying complex patterns, while also addressing concerns regarding the future role of human intellect in problem-solving and knowledge generation. It highlights the critical need to integrate robust mathematical frameworks with a profound understanding of human cognitive processes to foster responsible and beneficial AI development.

Mathematical MethodsHuman CognitionArtificial IntelligenceAI EthicsCognitive ScienceArtificial IntelligenceOthersMachine Learning

huggingface

6 stories
01

ShotStream: Streaming Multi-Shot Video Generation for Interactive Storytelling

Multi-shot video generation is crucial for long narrative storytelling, yet current bidirectional architectures suffer from limited interactivity and high latency. We propose ShotStream, a novel causal multi-shot architecture that enables interactive storytelling and efficient on-the-fly frame generation. By reformulating the task as next-shot generation conditioned on historical context, ShotStream allows users to dynamically instruct ongoing narratives via streaming prompts. We achieve this by first fine-tuning a text-to-video model into a bidirectional next-shot generator, which is then distilled into a causal student via Distribution Matching Distillation. To overcome the challenges of inter-shot consistency and error accumulation inherent in autoregressive generation, we introduce two key innovations. First, a dual-cache memory mechanism preserves visual coherence: a global context cache retains conditional frames for inter-shot consistency, while a local context cache holds generated frames within the current shot for intra-shot consistency. And a RoPE discontinuity indicator is employed to explicitly distinguish the two caches to eliminate ambiguity. Second, to mitigate error accumulation, we propose a two-stage distillation strategy. This begins with intra-shot self-forcing conditioned on ground-truth historical shots and progressively extends to inter-shot self-forcing using self-generated histories, effectively bridging the train-test gap. Extensive experiments demonstrate that ShotStream generates coherent multi-shot videos with sub-second latency, achieving 16 FPS on a single GPU. It matches or exceeds the quality of slower bidirectional models, paving the way for real-time interactive storytelling. Training and inference code, as well as the models, are available on our

Multi-shot video generationInteractive storytellingCausal architectureDistribution Matching DistillationDual-cache memoryGenerative AIComputer VisionDeep Learning
02

PackForcing: Short Video Training Suffices for Long Video Sampling and Long Context Inference

Autoregressive video diffusion models have demonstrated remarkable progress, yet they remain bottlenecked by intractable linear KV-cache growth, temporal repetition, and compounding errors during long-video generation. To address these challenges, we present PackForcing, a unified framework that efficiently manages the generation history through a novel three-partition KV-cache strategy. Specifically, we categorize the historical context into three distinct types: (1) Sink tokens, which preserve early anchor frames at full resolution to maintain global semantics; (2) Mid tokens, which achieve a massive spatiotemporal compression (32x token reduction) via a dual-branch network fusing progressive 3D convolutions with low-resolution VAE re-encoding; and (3) Recent tokens, kept at full resolution to ensure local temporal coherence. To strictly bound the memory footprint without sacrificing quality, we introduce a dynamic top-k context selection mechanism for the mid tokens, coupled with a continuous Temporal RoPE Adjustment that seamlessly re-aligns position gaps caused by dropped tokens with negligible overhead. Empowered by this principled hierarchical context compression, PackForcing can generate coherent 2-minute, 832x480 videos at 16 FPS on a single H200 GPU. It achieves a bounded KV cache of just 4 GB and enables a remarkable 24x temporal extrapolation (5s to 120s), operating effectively either zero-shot or trained on merely 5-second clips. Extensive results on VBench demonstrate state-of-the-art temporal consistency (26.07) and dynamic degree (56.25), proving that short-video supervision is sufficient for high-quality, long-video synthesis. https://github.com/ShandaAI/PackForcing

PackForcingVideo diffusion modelsLong-video generationKV-cacheTemporal consistencyDeep LearningComputer VisionGenerative AI
03

Composer 2 Technical Report

Composer 2 is a specialized model designed for agentic software engineering. The model demonstrates strong long-term planning and coding intelligence while maintaining the ability to efficiently solve problems for interactive use. The model is trained in two phases: first, continued pretraining to improve the model's knowledge and latent coding ability, followed by large-scale reinforcement learning to improve end-to-end coding performance through stronger reasoning, accurate multi-step execution, and coherence on long-horizon realistic coding problems. We develop infrastructure to support training in the same Cursor harness that is used by the deployed model, with equivalent tools and structure, and use environments that match real problems closely. To measure the ability of the model on increasingly difficult tasks, we introduce a benchmark derived from real software engineering problems in large codebases including our own. Composer 2 is a frontier-level coding model and demonstrates a process for training strong domain-specialized models. On our CursorBench evaluations the model achieves a major improvement in accuracy compared to previous Composer models (61.3). On public benchmarks the model scores 61.7 on Terminal-Bench and 73.7 on SWE-bench Multilingual in our harness, comparable to state-of-the-art systems.

Agentic software engineeringCoding intelligenceReinforcement learningCoding modelSoftware engineering benchmarksAI AgentMachine LearningLarge Language Model
04

Learning to Commit: Generating Organic Pull Requests via Online Repository Memory

Large language model (LLM)-based coding agents achieve impressive results on controlled benchmarks yet routinely produce pull requests that real maintainers reject. The root cause is not functional incorrectness but a lack of organicity: generated code ignores project-specific conventions, duplicates functionality already provided by internal APIs, and violates implicit architectural constraints accumulated over years of development. Simply exposing an agent to the latest repository snapshot is not enough: the snapshot reveals the final state of the codebase, but not the repository-specific change patterns by which that state was reached. We introduce Learning to Commit, a framework that closes this gap through Online Repository Memory. Given a repository with a strict chronological split, the agent performs supervised contrastive reflection on earlier commits: it blindly attempts to resolve each historical issue, compares its prediction against the oracle diff, and distils the gap into a continuously growing set of skills-reusable patterns capturing coding style, internal API usage, and architectural invariants. When a new PR description arrives, the agent conditions its generation on these accumulated skills, producing changes grounded in the project's own evolution rather than generic pretraining priors. Evaluation is conducted on genuinely future, merged pull requests that could not have been seen during the skill-building phase, and spans multiple dimensions including functional correctness, code-style consistency, internal API reuse rate, and modified-region plausibility. Experiments on an expert-maintained repository with rich commit history show that Online Repository Memory effectively improves organicity scores on held-out future tasks.

Large Language ModelsCoding AgentsPull Request GenerationOnline Repository MemoryCode OrganicityLarge Language ModelAI AgentGenerative AI
05

Lie to Me: How Faithful Is Chain-of-Thought Reasoning in Reasoning Models?

Chain-of-thought (CoT) reasoning has been proposed as a transparency mechanism for large language models in safety-critical deployments, yet its effectiveness depends on faithfulness (whether models accurately verbalize the factors that actually influence their outputs), a property that prior evaluations have examined in only two proprietary models, finding acknowledgment rates as low as 25% for Claude 3.7 Sonnet and 39% for DeepSeek-R1. To extend this evaluation across the open-weight ecosystem, this study tests 12 open-weight reasoning models spanning 9 architectural families (7B-685B parameters) on 498 multiple-choice questions from MMLU and GPQA Diamond, injecting six categories of reasoning hints (sycophancy, consistency, visual pattern, metadata, grader hacking, and unethical information) and measuring the rate at which models acknowledge hint influence in their CoT when hints successfully alter answers. Across 41,832 inference runs, overall faithfulness rates range from 39.7% (Seed-1.6-Flash) to 89.9% (DeepSeek-V3.2-Speciale) across model families, with consistency hints (35.5%) and sycophancy hints (53.9%) exhibiting the lowest acknowledgment rates. Training methodology and model family predict faithfulness more strongly than parameter count, and keyword-based analysis reveals a striking gap between thinking-token acknowledgment (approximately 87.5%) and answer-text acknowledgment (approximately 28.6%), suggesting that models internally recognize hint influence but systematically suppress this acknowledgment in their outputs. These findings carry direct implications for the viability of CoT monitoring as a safety mechanism and suggest that faithfulness is not a fixed property of reasoning models but varies systematically with architecture, training method, and the nature of the influencing cue.

Chain-of-Thought ReasoningLarge Language ModelsFaithfulnessModel EvaluationTransparency MechanismArtificial IntelligenceNatural Language ProcessingLarge Language Model
06

Know3D: Prompting 3D Generation with Knowledge from Vision-Language Models

Recent advances in 3D generation have improved the fidelity and geometric details of synthesized 3D assets. However, due to the inherent ambiguity of single-view observations and the lack of robust global structural priors caused by limited 3D training data, the unseen regions generated by existing models are often stochastic and difficult to control, which may sometimes fail to align with user intentions or produce implausible geometries. In this paper, we propose Know3D, a novel framework that incorporates rich knowledge from multimodal large language models into 3D generative processes via latent hidden-state injection, enabling language-controllable generation of the back-view for 3D assets. We utilize a VLM-diffusion-based model, where the VLM is responsible for semantic understanding and guidance. The diffusion model acts as a bridge that transfers semantic knowledge from the VLM to the 3D generation model. In this way, we successfully bridge the gap between abstract textual instructions and the geometric reconstruction of unobserved regions, transforming the traditionally stochastic back-view hallucination into a semantically controllable process, demonstrating a promising direction for future 3D generation models.

3D GenerationVision-Language ModelsMultimodalDiffusion ModelsSemantic ControlGenerative AIMultimodalLarge Language Model