Stories from Hacker News

The Hacker News Guidelines outline the community's expectations for submissions and discourse. It emphasizes intellectual curiosity, original sources, and substantive comments while prohibiting promotional content, AI reviews, and aggressive behavior. The goal is to maintain a high-quality, human-led environment focused on learning and respectful disagreement rather than ideological battles or shallow dismissals.

Mitchell Hashimoto, a long-time GitHub user and creator of Ghostty, announced that the project will leave GitHub due to recurring infrastructure outages that hinder productivity. Citing frequent downtime for pull requests and GitHub Actions, Hashimoto is planning to migrate independently while maintaining a read-only mirror, emphasizing that the decision stems from a desire to prioritize reliable software development.

The US government ordered Anthropic to suspend access to Fable 5 and Mythos 5, citing undisclosed national security concerns and potential jailbreak vulnerabilities. Anthropic defends its defense-in-depth strategy, noting these vulnerabilities are minor and common in other models. The company is complying while publicly criticizing the lack of transparency in the government's directive.

GPT-6 Astra demonstrates “Weight-Space Programming”: directly reading and editing neural-network weights without gradients or training loops. It inferred tasks and internal mechanisms from small networks, wrote game-playing networks that often outperformed PPO, and edited 0.016% of Qwen3.5-0.8B’s weights to reduce reasoning loops, raising MATH-500 from 7.9% to 48.1% and IFEval from 21.0% to 62.3%. Results remain limited: Astra could not improve Gemma’s ordinary reasoning errors or reliably detect sycophancy from weights alone.

The release of macOS Tahoe has introduced a significant design change regarding the corner radius of application windows. Many users find the new, larger radius aesthetically unappealing, but more importantly, it introduces a functional usability regression. The increased curvature reduces the effective hit area for window resizing by moving the 19 x 19 pixel target zone mostly outside the visible window frame. Analysis shows that about 75% of this target area now resides in empty space. Consequently, users attempting to grab the window corner from the inside frequently fail, as the system now prioritizes a gesture that begins outside the window's visual boundaries, making the interaction feel unnatural and error-prone compared to previous versions of the operating system.

HackerTube is a YouTube-style site built by Opus 5.5 from videos recommended on Hacker News. It processed all Hacker News comments, filtered spam and low-quality material with Sonnet, and identified about 40,000 comments recommending 6,454 YouTube videos across 150 topics from 2007–2026. The site offers curated recommendations without accounts, browser-session state, 50% thumbnails, keyboard shortcuts, and data export for creating a custom version.

Anthropic CEO Dario Amodei detailed the company's commitment to US national security while maintaining two critical ethical safeguards. Anthropic supports military AI applications like intelligence and cyber operations but refuses to permit use for mass domestic surveillance or fully autonomous weapons, despite Department of War threats to label the company a supply chain risk.

Current and former employees of Google and OpenAI have launched an open-source initiative and letter voicing concerns over the misuse of AI. The platform emphasizes secure, verified signing processes, including anonymous options, to build a broad coalition. The organizers aim to address potential risks while remaining independent of political parties or organizations.

We have launched Claude Fable 5, a state-of-the-art Mythos-class AI model featuring advanced capabilities in software engineering, research, and vision. To ensure safety, we implemented robust classifiers that route sensitive cybersecurity or biology queries to Claude Opus 4.8. Additionally, Claude Mythos 5 allows trusted partners access to the model without certain safeguards via a specialized program.

Anthropic has announced the release of Claude Opus 4.6, its most advanced AI model to date, featuring significant enhancements in coding, reasoning, and autonomous task execution. A major highlight is the introduction of a 1M token context window and adaptive thinking capabilities, which allow the model to adjust its reasoning depth based on task complexity. Claude Opus 4.6 excels in agentic workflows, outperforming competitors like GPT-5.2 in financial, legal, and multidisciplinary evaluations such as Terminal-Bench 2.0 and Humanity's Last Exam. New product integrations include Claude in Excel and a research preview for PowerPoint, alongside a multi-agent team feature in Claude Code. Despite these intelligence gains, Anthropic emphasizes a robust safety profile, including improved alignment and specialized cybersecurity safeguards to prevent potential misuse while maintaining the same pricing structure.

Bose has officially released the API documentation for its SoundTouch speaker line, offering a lifeline to users facing the impending end-of-life (EoL) status of these premium devices. Originally, Bose announced that these systems would lose cloud connectivity and app support by February 2025, effectively bricking several multi-room features. However, the company has clarified that Spotify Connect and AirPlay 2 will remain functional, allowing continued wireless streaming. Furthermore, a local version of the SoundTouch app will be released in 2026 to facilitate control without cloud dependency. This move addresses significant customer backlash regarding the planned obsolescence of expensive hardware, ensuring that these high-end speakers retain functionality through local integration and open documentation even after official manufacturer support ceases.

DeepInfer is an open-source, high-performance offline inference framework for serving DeepSeek models at large scale, currently supporting DeepSeek-V4.1-Flash. It provides optimized kernels and runtimes, Attention DP, balanced CP, MoE EP + EPLB, DSpark verification, SWA bounded replay, asynchronous scheduling, batch-invariant and deterministic inference, and structured outputs. It runs on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs, but a complete service still requires KV cache, disaggregated execution, and replica load balancing.