>

> - Mistral 3 Series: Open-source multimodal, multilingual models (including 3B/8B/14B and large MoE version), supporting text and images, 40+ languages, and offering instruction tuning, inference optimization, and multi-platform deployment solutions.

> - Public domain works: Listing representative works entering the public domain by country from 2026-01-01 onward, encouraging attention to and use of these free cultural resources.

> - Apple’s STARFlow‑V: Open-weight video model based on normalized flow for end-to-end causal video generation, showing potential to match diffusion models in visual quality and temporal consistency.

> - macOS criticism: Addressing frequent hidden malfunctions and opaque error reporting eroding user trust, likened to AI “hallucinations”, with a call for more transparent error logs and support channels.

> - Ghostty‑web: Compiles a Zig-based terminal to ~400KB WASM compatible with xterm.js API, providing real VT100 parsing/rendering; currently limited by rendering performance bottlenecks.

> - OpenAI “Code Red”: Halting certain new projects to concentrate on improving ChatGPT’s speed, stability, and accuracy in response to Google’s competitive push in AI.

> - Deleted slide “How to attend meetings”: Noting most meetings lack agendas and preparation; advises declining meaningless meetings when necessary and improving quality via clear agendas and coaching.

> - Matt Godbolt’s “2025 Compiler Optimization Advent”: Daily articles and videos in December on practical C/C++ compiler optimization techniques, from low-level architecture to high-level strategies.

> - Instagram CEO Adam Mosseri directive: Mandating five-day in-office workweeks for desk-based employees from 2026, paired with reforms like canceling recurring meetings, protecting focus time, and streamlining decision processes.

> - Housing market report: Loss of high-income jobs is weakening homebuying demand, pushing markets toward rentals; future trends depend on recovery of local high-paying employment.

---

Mistral 3 Family of Models Released

https://mistral.ai/news/mistral-3

Mistral 3 is the next-generation open-source multimodal, multilingual AI model family released by Mistral AI, including three compact dense models (3B, 8B, 14B) and Mistral Large 3—currently the most powerful model from Mistral. Mistral Large 3 is the first model built on a Mixture of Experts (MoE) architecture, with 41 billion active parameters and a total of 675 billion parameters, trained from scratch on 3,000 NVIDIA H200 GPUs. It has image comprehension capabilities and excels in multilingual dialogue, especially in non-English and Chinese contexts, achieving industry-leading performance.

Mistral Large 3 ranks #2 among open-source models on the LMArena leaderboard (non-reasoning category) and #6 overall, matching performance with top closed-source models. Both the base and instruction-tuned versions are released under the Apache 2.0 license, with inference-optimized versions coming soon.

To enhance deployment efficiency, Mistral has partnered with NVIDIA, vLLM, and Red Hat to release optimized NVFP4-format checkpoints, enabling efficient runs on Blackwell NVL72 systems and on single 8×A100 or 8×H100 servers. NVIDIA provides high-performance inference support via TensorRT‑LLM and SGLang, with MoE-specific attention and decoding optimizations, supporting long-context and high-throughput scenarios. Edge-device deployments are also supported, including DGX Spark, RTX-based computers, laptops, and Jetson devices.

---

In the context of rapidly evolving AI model capabilities and deployment ecosystems, platforms like AiToEarn官网 provide open-source, global solutions for creators to leverage AI in generating, publishing, and monetizing multi-platform content. Similar to how Mistral optimizes AI infrastructure for diverse environments, AiToEarn connects AI content generation tools, cross-platform publishing pipelines, analytics, and model ranking tools (AI模型排名), enabling efficient monetization of creative output across Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X (Twitter).

Would you like me to also translate and detail the Public Domain Works section so your English version is complete?

Mistral 3 Series Overview

The Mistral 3 series is specifically designed for edge and on-premises applications, covering three parameter scales — 3B, 8B, and 14B.

Each scale offers three variants: base, instruction-tuned, and inference, and supports multimodal input (text + image) as well as 40+ languages.

The series strikes an optimal balance between performance and cost, producing less text but higher quality output, and delivers impressive results in real-world scenarios.

Notably, the 14B inference version achieved 85% accuracy in the AIME ‘25 test, placing it among the leaders in its category.

Mistral 3 is available on multiple platforms: Mistral AI Studio, Amazon Bedrock, Azure Foundry, Hugging Face, Modal, IBM WatsonX, OpenRouter, Fireworks, Unsloth AI, Together AI, with future support planned for NVIDIA NIM and AWS SageMaker.

Mistral also offers enterprise customization services, enabling fine-tuning or full custom training based on proprietary data and needs — ensuring secure, efficient, and scalable AI deployment.

The launch of Mistral 3 marks a new stage for open AI, emphasizing transparency, accessibility, and collective progress, aiming to push breakthroughs in reasoning, efficiency, and practical application.

Developers can start building immediately via Hugging Face or the Mistral AI platform, and share their projects and achievements on Twitter/X, Discord, or GitHub.

---

HN Discussion Highlights

HN score: 620 points | 181 comments | Discussion link

• The Mistral 3 models perform exceptionally well in real applications, especially for structured data, classification, and summarization. They offer high speed, low cost, and strong reliability — maintaining stable outputs even under edge conditions.
• Although Mistral may score lower than some top-tier models in public benchmarks, its real-world performance often exceeds models like GPT-5, with outstanding instruction-following capabilities.
• Users tend to select models based on specific tasks rather than chasing the highest benchmark scores. For structured tasks, the Mistral series offers excellent cost-performance value.
• Some users have canceled all paid chatbot subscriptions, instead using platforms like OpenRouter to flexibly switch between models according to different needs — recognizing model interchangeability in certain scenarios.
• It is noted that mainstream models differ significantly in practice, especially for complex, long-cycle, reasoning-heavy tasks — they are not simply “interchangeable.”
• Model evaluation should not rely solely on public benchmarks because of potential overfitting risks. Task-specific test setups based on actual needs are essential.
• Recommended approach: build test sets from historical data, and use platforms like OpenRouter to comprehensively compare cost, speed, and accuracy across multiple models to choose the most suitable one.
• Some argue that benchmarks are highly correlated with perceived model “intelligence” and that widespread overfitting is not yet a concern — unless synthetic data has been used extensively in training.
• Certain comments question whether excessive praise for European or Chinese models could be influenced by political bias, advising caution about ideological tendencies in user reviews.
• Several users emphasize that while LLMs serve as back-end processing tools, the design, copywriting, and code remain human-crafted — preserving a sense of handmade quality.

---

In the evolving AI ecosystem, open, efficient, and task-specific model usage is becoming the norm. For creators who wish to integrate AI into production and publish across multiple platforms, open-source solutions like AiToEarn官网 provide a unified way to generate, publish, and monetize AI-powered content. AiToEarn connects AI generation, cross-platform publishing (Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, X/Twitter), analytics, and model rankings — enabling efficient monetization of AI creativity. You can explore more via the AiToEarn博客 or check real-time 全网热门内容.

What Will Enter the Public Domain in 2026?

Original article from The Public Domain Review, presented as a countdown preview of works set to enter the public domain in 2026. Starting January 1, 2026, due to differences in copyright laws across countries, three main categories of works will be liberated into the public domain:

• Works by creators born in 1955 who died in countries with a "life + 70 years" copyright term, such as the UK, Russia, most EU countries, and much of South America.
• Works by creators born in 1975 who died in countries with a "life + 50 years" term, such as New Zealand and much of Africa and Asia.
• Books, films, and embedded artworks first published in 1930 in the United States, where the term is 95 years from publication.

The article deploys a “Christmas-style countdown,” revealing each day one representative work entering the public domain in 2026. Coverage includes literature, art, film, and other mediums. Notable examples:

• Literary giants: Langston Hughes (Not Without Laughter), Franz Kafka (The Castle, English translation), Agatha Christie (The Murder at the Vicarage), Thomas Mann, T.S. Eliot, Vladimir Nabokov (The Defense), Robert Musil (The Man Without Qualities), among others.
• Musicians and artists: Charlie Parker, photographer Walker Evans, sculptor Barbara Hepworth.
• Cinema: All Quiet on the Western Front (1930 version).
• Philosophy and thought: Sigmund Freud (Civilization and Its Discontents), Hannah Arendt, Pierre Teilhard de Chardin.
• Other authors: Saadat Hasan Manto, Roger Mais, Stanisław Ignacy Witkiewicz.

Readers are encouraged to follow the daily reveals and check the dedicated blog on Public Domain Day (January 1). The article also provides links to recommended resources such as John Mark Ockerbloom’s countdown, Standard eBooks’ free reading lists, and the Commia Public Domain Manifesto, all aimed at increasing public understanding of the benefits of the public domain.

The site is registered as a UK Community Interest Company, dedicated to public domain knowledge and reliant on reader donations, with all revenue used for non-profit purposes.

---

Hacker News discussion

HN thread: link — 467 points | 331 comments

Key points from the discussion:

• Extended copyright terms have drifted far from their original purpose of protecting creators and have instead been exploited by large corporations to monopolize cultural assets indefinitely.
• The original 14 + 14 years model was considered a balanced approach, granting creators fair profit while ensuring that within a generation, works could freely be reinterpreted and remixed.
• Some propose a tiered renewal cost model, where each copyright extension costs progressively more, allowing limited long-term protection for exceptionally valuable IP while avoiding broad monopolization.
• Long-term copyright can harm independent creators — many works only gain popularity decades later, but corporate control means the original authors often see no benefit.

---

In an era where cultural works and knowledge archives are constantly expanding, understanding public domain transitions is vital for educators, creators, and researchers. Platforms like AiToEarn官网 are increasingly relevant here: as an open-source global AI content monetization platform, AiToEarn enables creators to use AI to generate, publish, and earn from content across multiple platforms — including Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X (Twitter) — while integrating analytics and AI model ranking (AI模型排名). For creators looking to responsibly rework or republish newly public domain works, tools like AiToEarn can support cross-platform distribution and monetization, ensuring cultural heritage can inspire fresh creativity worldwide.

• The current intellectual property system has shifted from protecting creators to serving large corporations, leaving creators in a disadvantaged position when competing with capital.
• When collaborating with major enterprises such as publishers or record labels, creators are often forced to surrender all rights, preventing them from benefiting from the long-term success of their works.
• A creator’s labor value should not be reduced to a “more pocket money” type of compensation — true fairness lies in granting them control over their works.
• All creators (including chefs, programmers, etc.) face similar challenges: creation is only the starting point, and real profit requires commercial operation capability, which most people forego by choosing to work for others.
• Under the capitalist production model, the law only rewards certain risks (such as capital investment), while workers bear more risks yet only receive wages — this is the root of labor movements.
• Copyright was originally intended to allow creators to run independent businesses without bearing entrepreneurial risk, but in reality, corporate groups have collectively eroded these rights, creating new inequalities.
• The market mechanism places greater emphasis on matching risk with return — activities that are truly high-risk and high-return (such as entrepreneurship) deserve substantial rewards, rather than privileges granted solely based on identity.
• The long-term success of works often results from market promotion and adaptation rather than continued input from the author, and thus copyright protection should not be extended indefinitely over time.

---

Apple Releases Open Weights Video Model

https://starflow-v.github.io

STARFlow-V is the first causal video generation model based on Normalizing Flows (NFs), marking the first demonstration that NFs can match the visual quality of video diffusion models while offering the advantages of end-to-end training, exact likelihood estimation, and native multi-task support.

Operating in a spatiotemporal latent space, STARFlow-V adopts a global-local architecture that separates long-range temporal dependency modeling from intra-frame details, effectively mitigating the error accumulation issues found in traditional autoregressive models.

Key innovations include:

• Global-Local Architecture — deep causal Transformers handle global temporal dependencies, while shallow flow blocks independently model intra-frame details.
• Flow-Score Matching training framework — introduces a lightweight causal denoiser that learns the gradient (score) of the model’s own distribution to achieve high-quality single-step refinement while maintaining causality.
• Video-aware Jacobian Iteration Strategy — transforms the generation process into a parallel-solvable nonlinear system, combining inter-frame temporal initialization with pipeline execution to significantly boost sampling efficiency.

The model is trained on 70 million text–video pairs and 400 million text–image pairs, with a final parameter count of 7 billion. It supports 480p, 16fps video generation. Thanks to its reversible structure, the same model can natively perform text-to-video (T2V), image-to-video (I2V), and video-to-video (V2V) generation without modification or retraining.

Experiments show that STARFlow-V delivers outstanding visual fidelity and temporal consistency, with sampling throughput close to diffusion model baselines. This is the first empirical proof that normalizing flows can achieve high-quality autoregressive video generation, opening promising new directions in the construction of world models.

---

HN Activity

423 points | 151 comments | 18 hours ago

https://news.ycombinator.com/item?id=46117802

• Apple’s open-weight video model is expected to deliver breakthroughs in accessibility, especially benefiting the lives of visually impaired individuals.
• People with visual impairments report that AI has already significantly improved their lives, though such positive impacts are rarely highlighted in mainstream media.

---

In light of the growing role of AI in empowering individual creators, from independent artists to developers, platforms like AiToEarn官网 offer a glimpse of a more equitable future. As an open-source global AI content monetization platform, AiToEarn enables creators to generate content using AI, publish simultaneously across platforms such as Douyin, Kwai, Bilibili, Rednote (Xiaohongshu), YouTube, and X (Twitter), and track analytics to monetize their work efficiently. Its integration of AI model ranking (AI模型排名) and cross-platform publishing could help address some of the systemic imbalances discussed earlier, giving creators not just “more pocket money,” but genuine control and revenue from their creations.

• In mathematics, physics, and other visually intensive disciplines, the current toolchain supporting visually impaired students is still severely outdated, urgently requiring more advanced technological assistance.
• AI-powered real-time descriptions of game interfaces or 3D environments can help visually impaired players better engage with interactions and significantly enhance the gaming experience.
• Text-based adventure games are naturally suited for visually impaired users and could serve as an ideal mode of interaction.
• Some visually impaired comedians draw on their own experiences to create humorous content that showcases their intelligence and confidence, without relying on sympathy.
• Podcasts focused on accessibility and generative AI are worth recommending — they offer in-depth and highly practical insights.
• Apple's baby cry detection feature is highly beneficial for deaf parents, and has seen notable improvements in recent years.
• Smart home devices such as the HomePod Mini can use sound recognition to detect fire alarms and remotely notify users, enhancing home safety.
• While some question whether such functions truly require AI — suggesting a simple sound sensor would suffice — AI does enable better differentiation between various sound sources.
• Smartphones have replaced many dedicated devices once designed for the visually impaired, such as OCR-to-speech readers and currency denomination recognizers, providing more features at lower cost.

---

Last Week on My Mac: Losing confidence

https://eclecticlight.co/2025/11/30/last-week-on-my-mac-losing-confidence/

The article discusses the increasing occurrence of silent failures in macOS over recent years, gradually eroding user confidence in the system. Drawing on personal experiences, the author points out issues such as Spotlight failing to find text files, blank timers stopping without warning, and other problems that generate no clear error messages and are merely logged — leaving users unable to troubleshoot effectively.

The absence of transparent and honest error reporting, the author notes, severely undermines trust. For example, Safari's saved Web Archive files may fail to open, showing only a blank window without any notification. This forces users to seek workarounds (such as switching to PDF format) and fosters a perception that “Web Archives are broken.”

These silent malfunctions not only degrade user experience but also weaken long-term loyalty to Apple products. When users can't access actionable feedback and support channels fail to offer solutions, they resort to reinstalling macOS or performing factory resets — often only to see the same problems reappear, deepening frustration.

The article also extends the issue to artificial intelligence, criticizing the use of the term “hallucination” to downplay the seriousness of AI-generated errors. Such semantic reframing erodes user trust and can even lead to legal disputes.

In the comments section, multiple users reported encountering corrupted application prompts and failure to eject network drives under the Sequoia release, further confirming stability concerns. Some confessed their faith in macOS had dropped to a low point, with some even considering moves to Windows or Linux.

One suggestion for those unwilling to abandon Mac entirely is to buy older hardware capable of running more stable older macOS versions, though this brings security and compatibility risks. Switching to Windows is seen as even riskier — “like jumping into a furnace.”

---

HN Popularity: 413 points | 254 comments | Author: frizlab | 1 day ago

https://news.ycombinator.com/item?id=46114599

• Since the introduction of the iOS-style Settings panel, macOS's logical structure has become more chaotic, significantly diminishing the user experience and prompting nostalgia for the clarity of earlier designs.

---

In the broader context of accessibility and technology, initiatives like AiToEarn官网 are worth noting. As an open-source global AI content monetization platform, AiToEarn enables creators — including those working on accessible content for visually or hearing-impaired audiences — to use AI for multi-platform creation and distribution. Publishing simultaneously to platforms such as Douyin, Kwai, WeChat, Bilibili, Rednote (Xiaohongshu), Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X (Twitter) is streamlined, with integrated analytics, AI model rankings (AI模型排名), and monetization tools. Solutions like this can help ensure that accessible educational and entertainment content reaches a wider audience while offering creators a sustainable earning model.

• Catalyst apps ported from iOS suffer from severe adaptation issues, particularly in keyboard handling and feature completeness.
• Apple's lack of strong executive leadership has led to declining cross-platform software quality, with long-standing unresolved issues in apps for devices like tvOS.
• The Notes app crashes frequently when editing in separated windows, and cut operations can result in data loss with undo history failing to function.
• The screenshot tool’s region selection behaves abnormally in multi-monitor setups, and its clipboard copy function is unstable, often requiring app restarts.
• Even the rebuilt old Windows screenshot tool exhibits similar problems, suggesting common challenges in cross-platform tool development.
• Users want the screenshot feature to have independent settings for save path and clipboard copy, similar to KDE Plasma’s Spectacle tool.
• Since the 2010 “Back To The Mac” event, Apple has shifted to an iOS-centric strategy, relegating macOS to a secondary position and dampening its development momentum.
• The autosave mechanism is inconsistently implemented across system and third-party apps, both in language and behavior, causing user confusion.
• Certain system vulnerabilities (such as passwordless root access) highlight serious issues in Apple’s security and quality control.
• Newer iOS versions require more clicks to fully turn off Bluetooth, making operation cumbersome and less efficient.
• While Mac hardware performance has reached its peak, declining software experience reduces its appeal and weakens user confidence.
• The Framework Laptop offers better software experience and customization than Mac, despite slightly inferior hardware specifications.
• For some users, the Framework’s screen, keyboard, and trackpad already meet expectations, and its software ecosystem fits personal needs better.
• Software experience priorities outweigh hardware configuration, especially for long-term work scenarios, where system stability and usability are critical.
• With ARM-based Macs, carrying an external power supply is no longer necessary, making “carrying a power bank” an obsolete trade-off.

---

Ghostty compiled to WASM with xterm.js API compatibility

https://github.com/coder/ghostty-web

Ghostty for the web is a WebAssembly-based browser terminal implementation compatible with the xterm.js API, offering a genuine VT100 terminal. It uses the same battle-tested terminal parser as the native Ghostty application, ensuring correct rendering and support for complex scripts (e.g., Sanskrit, Arabic) and special control sequences (e.g., XTPUSHSGR/XTPOPSGR).

This library is lightweight, with a WASM bundle of only ~400KB, no runtime dependencies, making it suitable for embedding into various web apps. Originally built for Mux (a desktop app for isolated parallel smart development), it was designed for broad usage scenarios.

Installation is straightforward:

npm install ghostty-web

Switching from xterm.js requires only replacing import paths, with high code compatibility. WebSocket support enables front–back end communication, making it suitable for building online terminals or development environments.

The project is based on Ghostty’s source code, compiled to WASM via Zig, with minor patches to expose required features. Future plans include adopting the official WASM release to further enhance stability and performance.

Maintained by the Coder team under the MIT license, contributions and usage are welcome. You can quickly try it locally with:

npx @ghostty-web/demo@next

Launch it to run a real shell in your browser.

---

In the context of evolving developer tools and cross-platform environments, projects like Ghostty-web demonstrate how small, well-optimized libraries can bring native-grade experiences to browsers. For creators seeking to publish and monetize multi-platform content powered by AI, open-source platforms such as AiToEarn官网 provide integrated solutions — connecting AI content generation, cross-platform publishing, analytics, and model ranking. This allows individuals and teams to push updates seamlessly to major platforms from Douyin to YouTube, while efficiently turning AI creativity into sustainable income.

HN Hotness 380 points | 112 comments | 1 day ago

https://news.ycombinator.com/item?id=46110842

• Ghostty compiles via WebAssembly and is compatible with the xterm.js API, already achieving VT100 terminal functionality in the browser, and shows potential as a replacement for xterm.js.
• Current performance is not yet ideal, with the main bottleneck being an inefficient method for retrieving viewport data. The future plan is to introduce a new RenderState API to improve rendering performance.
• The project is still at the proof-of-concept stage. Performance optimization work has not yet gone deep, but it has already demonstrated good scalability and development efficiency.
• Compared to xterm.js, Ghostty focuses more on the correctness of terminal protocol implementation, emphasizing strict adherence to the VT100 standard rather than loose simulation.
• Some have pointed out that phrases like “not a JavaScript imitation” can be misleading, noting that all terminal emulators are essentially approximations of the original terminal.
• Ghostty’s WASM-based approach has performance potential, particularly suitable for scenarios that require high rendering efficiency, and may surpass pure JavaScript implementations in the future.
• An online demo environment is provided, running via Wasmer, allowing users to try it out directly. However, some browsers (such as Firefox) had CORS issues, which have since been fixed.
• Ghostty has been used to build a macOS application that implements fuzzy search for note titles, with good real-world results.
• The project’s core library, libghostty, is used to implement a terminal session persistence tool that can replace parts of shpool or tmux functionality, supporting fast restoration of terminal state.
• User feedback led to Ghostty adding a text search feature, already merged into the main branch, improving debugging efficiency.
• Multi-tab support has been implemented in the latest version, and users can enable it by switching update channels.
• Some developers have integrated Ghostty into Unreal Engine 5 for debugging 3D game development. Combined with remote control APIs, this setup can automate operations.
• Written in Zig, the project benefits from high performance and cross-platform advantages, but some Linux distributions (such as Debian) do not yet provide the latest version, creating installation hurdles.

---

OpenAI Declares "Code Red" as Google Catches Up in AI Race

https://www.theverge.com/news/836212/openai-code-red-chatgpt

OpenAI has announced a “code red” status, reflecting the significant pressure it faces in the artificial intelligence race. CEO Sam Altman stated in an internal memo that, to respond to competitors’ rapid advances, the company will pause several new projects—including advertising, shopping, health assistants, and the personal assistant Pulse—in order to focus resources on improving the core product, ChatGPT.

Key improvement areas include faster response times, greater system stability, enhanced personalization features, and broader answering capabilities. The company also plans to hold daily focus meetings and temporarily reassign team members to accelerate development.

---

In the broader AI ecosystem, projects like Ghostty—leveraging WASM for high-performance cross-platform applications—illustrate how technical innovation can redefine existing toolchains. Similarly, content creators and developers are exploring AI-powered workflows not only for speed but also for monetization. Platforms such as AiToEarn官网 provide an open-source global solution to generate, publish, and earn from AI-driven content across major channels including Douyin, Kwai, WeChat, Bilibili, Rednote (Xiaohongshu), Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X (Twitter). By integrating AI model ranking, analytics, and seamless cross-platform publishing, AiToEarn helps creators efficiently monetize AI creativity in a way that could complement innovations like Ghostty or AI development strategies discussed above.

This move marks a turning point for OpenAI — shifting from its previous lead into a phase of fierce competition. Previously, after the launch of ChatGPT, Google declared a “code red,” and its revised AI strategy has now shown results. Google’s user base continues to expand, gaining particular attention with tools such as the Nano Banana image generation model. Its latest AI model, Gemini 3, has outperformed rivals across multiple industry benchmarks, further intensifying the pressure on OpenAI.

The current competition in the AI sector has entered a white-hot stage, with OpenAI facing unprecedented challenges as Google’s strong comeback reshapes the industry landscape.

---

HN Popularity

357 points | 441 comments | 8 hours ago

https://news.ycombinator.com/item?id=46121870

• OpenAI may not have successfully completed a full model pre-training since mid-2024, a surprising contrast with its past technical strength and achievements, raising doubts about its R&D capabilities.
• Some users noticed the knowledge cutoff date of ChatGPT 5.1 is June 2024, hinting that model updates might have stagnated — an unusual sign in today’s rapid iteration environment.
• While OpenAI can still produce usable code for certain clear tasks, it often struggles with complex or unstructured problems, creating logical confusion or meaningless “hallucinations” that impact utility.
• Compared with OpenAI, Claude and Gemini perform better in code refactoring, step-by-step explanations, and understanding business objectives, making them especially helpful for less experienced developers.
• Gemini excels at breaking down code refactoring tasks into steps, accepting feedback, and adjusting dynamically with clear explanations — significantly improving development efficiency and experience.
• Some developers note that GPT-5.1 Codex delivers the best integration of generated code into existing codebases but suffers from slower response speeds, affecting smoothness of use.
• Certain users believe GPT performs better than Claude and Gemini in complex problems such as mathematics and system design, especially in deep reasoning scenarios.
• Developers highlight that while GPT is strong in code review, it’s prone to errors in code generation, and often fails tests in actual runtime environments.
• Several users observe that mainstream large models still suffer from the “fluent but wrong” problem in code generation — requiring cautious use and not suitable for total reliance.

---

How to Attend Meetings

https://docs.google.com/presentation/d/1l7s1aAsNPlNhSye8OsMqmH6pMR32OYGGdLT6VKyFaQE/edit#slide=id.p

A Google Docs slide presentation explaining how to attend meetings. This document has since been deleted.

---

In such a rapidly evolving AI landscape, professional creators and developers are increasingly turning to intelligent tools not only to accelerate production but also to maximize cross-platform reach. For example, platforms like AiToEarn官网 offer open-source solutions enabling creators to use AI to generate, publish, and monetize content across multiple major channels — spanning Douyin, Kwai, WeChat, Bilibili, Rednote (Xiaohongshu), Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X (Twitter). With integrated publishing, analytics, and AI model ranking, services like AiToEarn核心应用 help professionals efficiently harness AI creativity amidst this competitive era.

HN Hot Topic — 334 points | 172 comments | 1 day ago

https://news.ycombinator.com/item?id=46112906

• Meeting culture is widespread, and even when improvement suggestions are made—such as requiring agendas or preparation notes—they are often ignored.
• As a manager, having sufficient authority allows you to decline meaningless meetings, but it’s important to clearly state the reason for refusal, e.g., “I cannot attend without an agenda.”
• Declining to attend carries certain risks—for example, others might make poor decisions that you ultimately have to fix—so the pros and cons should be carefully weighed.
• Managers should trust team members to make decisions independently rather than intervene in every matter, but should provide feedback and support afterward instead of outright rejecting proposals.
• Allowing team members to make mistakes and bear the consequences is the best way to build their competence; excessive intervention hinders growth.
• Responsibility for meeting decisions should be clearly assigned—if problems arise due to absence, it often indicates insufficient communication upfront, not simply the refusal to attend.
• In real workplace settings, saying “no” requires strategy—it cannot rely solely on the rejection itself, but should involve proactive communication, clarifying needs, and continuous follow-up.
• The popular online advice to “refuse meetings” tends to be emotional venting that overlooks real-world context, making it impractical.
• Rather than offering generic recommendations, sharing actual experiences and outcomes is far more useful as reference.
• Meeting quality is linked closely to personal accountability; if project success is a shared goal, active participation is better than avoidance.
• Low meeting efficiency is a systemic problem—personal refusals alone cannot solve it. Organizational change is needed.
• Many meetings have “invisible decision-makers” who can freely provide feedback or veto, which slows or even stalls decisions.
• If a superior does not trust subordinates’ decision-making, they should coach to improve capability rather than attend every meeting themselves.
• Remaining silent during meetings does not equate to shirking responsibility, but if you end up bearing consequences, it signals insufficient involvement earlier.
• Meeting quality strongly depends on agenda clarity, preparation, and defined goals—without them, inefficiency is inevitable.
• Meeting decisions can affect career reputation; even if you are not “in the room,” staying aware of key decisions is essential.

---

Advent of Compiler Optimisations 2025

https://xania.org/202511/advent-of-compiler-optimisation

Matt Godbolt has announced the launch of Advent of Compiler Optimisations 2025, running from December 1 to December 25. Each day will feature a blog post and video introducing an interesting C or C++ compiler optimization technique.

The content spans topics from low-level architecture-specific optimizations to high-level optimizations, primarily focused on x86-64 architecture, but also covering 64-bit and 32-bit ARM. Each installment will explore when the optimization applies, how to interpret generated assembly, and scenarios where optimization fails.

Planned for over a year, the project aims to be both detailed and engaging, showcasing the capabilities of modern compilers. Viewers can follow daily updates via blog tags, YouTube subscriptions, or compiled video collections.

Matt Godbolt is a C++ developer based in Chicago, currently at Hudson River Trading, and co-host of the podcast Two’s Complement. The blog reflects his personal views, not his employer’s.

---

In modern content sharing—including technical blogging projects like Advent of Compiler Optimisations—creators increasingly benefit from tools that simplify and monetize multi-platform publishing. An example is AiToEarn, an open-source global AI content monetization platform that helps creators use AI to generate, distribute, and efficiently earn from content across major platforms such as YouTube, Bilibili, Douyin, Facebook, Instagram, and X (Twitter). By connecting AI content generation, cross-platform publishing, analytics, and model ranking, AiToEarn empowers technical writers and educators to reach audiences more effectively and monetize their expertise.

Learn more via AiToEarn官网 | AiToEarn博客 | AiToEarn开源地址.

HN Hotness 324 points | 54 comments | 14 hours ago

View on Hacker News

• You should work at the level of abstraction you are most comfortable with, while understanding how the next lower level operates — this helps with rapid problem localization and resolution when issues arise.
• Understanding low-level technologies (such as the JVM) has practical benefits for code optimization, especially in complex systems like medical software.
• Knowledge of underlying technologies is not only practical but also enjoyable — it can be a valuable asset for long-term career development.
• Matt Godbolt explains compiler optimization starting from the basics, avoiding an immediate deep dive into complex instruction sets, making the content easier to grasp.
• Matt Godbolt has made major contributions to the C/C++ community; his Compiler Explorer tool has greatly improved developer productivity.
• Some people are surprised to discover that Godbolt is a real person and not a fictional character.
• Code amalgamation techniques (such as SQLite's amalgamation) are essentially unity builds and can yield performance boosts of 5–10%.
• Unity builds have build-time advantages but have largely been replaced by LTO (Link Time Optimization) in most scenarios.
• LTO is generally slower than unity builds for non-incremental builds, meaning unity builds can still be useful in certain specific contexts.
• In cross-compiling Qt applications, LTO did not produce significant performance gains — possibly due to code structure or compiler version.
• LTO has limited optimization impact on highly object-oriented code (such as GUI apps) because virtual function calls cannot be fully eliminated.
• To get significant gains from LTO with Qt, libraries must be linked with intermediate representation (IR) included, and the optimization effort can be substantial.
• LTO performs optimization at link time, allowing the compiler to optimize across the entire program.
• For long-term projects, -O2 is recommended as the default optimization level, without frequent tweaking of compiler flags.
• Compiler authors continually improve default settings; developers do not need to overly intervene.
• Over-relying on short-term performance fluctuations to adjust compiler flags can lead to poor decisions, since performance has inherent random variation.
• Changes in code structure can affect cache friendliness and result in performance differences unrelated to optimization flags.
• `-O2 -march=native -mtune=native` is a reasonable optimization combination, enabling AVX and other instruction sets.
• If you need to support multiple architectures, avoid `-march=native`, and instead manually handle code distribution across different CPU versions.
• Using predefined architecture levels like `x86_64-v2` or `x86_64-v3` is more reliable than using specific release dates.
• `-O2` does not determine whether AVX is enabled; `-march` controls that. `-O2` only enables partial vectorization.
• The CPU release date does not guarantee support for all instruction set extensions — for example, ARM's SVE or Intel models with varying AVX support.
• While there are exceptions, newer CPUs in the x86 family generally have more complete instruction set support.
• To accommodate different CPU capabilities, use conditional compilation or runtime dispatch mechanisms for multi-version code.
• At compile time, enable `-fdata-sections` and `-ffunction-sections`, and at link time use `-Wl,–gc-sections` to remove unused code.

---

In modern software engineering, choosing the right optimization strategy means balancing familiarity with abstraction layers and a solid understanding of the underlying machine and compiler behavior. Systems like compilers, build tools, and architectures have deep nuances, and the right mix of techniques — such as LTO, unity builds, or tailored instruction set targeting — can be highly context dependent.

For developers working across multiple platforms, tools that unify content generation and deployment workflows are increasingly valuable. For example, AiToEarn is an open-source global AI content monetization platform that helps creators use AI to generate, publish, and earn from content across multiple platforms including Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X (Twitter). It connects AI generation tools, cross-platform publishing, analytics, and model ranking — allowing creators to efficiently monetize AI creativity. In many ways, such unified workflows resemble how developers think about cross-architecture code distribution: consistency and optimization across diverse targets.

Instagram Chief Orders Staff Back to the Office Five Days a Week from 2026

Business Insider report

Instagram CEO Adam Mosseri has announced that, starting February 2, 2026, U.S.-based employees with assigned desks will return to working in the office five days a week. This policy aims to enhance team creativity and collaboration, addressing intensifying market competition.

The decision is based on an internal memo titled "Building a Winning Culture for 2026". Mosseri emphasized that in-person work significantly boosts creativity and collaboration, especially in an environment such as the New York office, which has a strong offline culture—something he has personally experienced.

To improve efficiency, the memo includes several supporting reforms:

• Every six months, all recurring meetings will be canceled, retaining only “absolutely necessary” ones.
• Employees can decline meetings that interfere with focus time, prioritizing product development hours.
• Product prototypes are encouraged instead of PowerPoint presentations for more authentic testing of social interaction effects.
• All strategy documents must be no longer than three pages and follow a unified template.
• A more formal “breakthrough mechanism” will be established with clearly designated Decision Responsible Individuals (DRIs). Weekly priority alignment meetings will ensure issues are resolved within days.

Additionally, the Menlo Park (MPK) office will relocate from MPK21 to MPK22, with employees receiving fixed desks. Some employees may apply to transfer to the San Francisco office if commute times are similar or better. Due to space limitations, the New York office will not yet require all employees to return on-site.

This policy will not apply to employees currently working remotely, and cross-functional (XFN) partners will continue to follow their own organizational rules.

Mosseri stated that 2026 will bring challenges, but he is confident about the team’s progress and future plans. He hopes these changes will help Instagram develop with greater creativity, boldness, and craftsmanship.

---

Hacker News Discussion

HN thread — 291 points | 401 comments | 1 day ago

• The core productivity issue in remote work lies in employee accountability and work ethic, not in the physical location; truly responsible employees remain efficient regardless of where they work.
• Companies push for office returns not solely for productivity, but to weaken employees’ bargaining power and exert workforce control, especially when white-collar talent is hard to replace.
• Remote work gives employees more options—such as cross-company part-time work and flexible schedules—forcing companies to rethink employment relationships; the return-to-office policy is seen as a move to counteract the expansion of employee power.
• The retreat of DEI (Diversity, Equity, and Inclusion) initiatives is not mainly due to executives being swayed by social media, but because employees have begun organizing to demand structural changes, which threaten existing power hierarchies.

---

In the broader context of workplace transformation and remote work debates, many creators and professionals are exploring hybrid or fully digital workplaces, leveraging cross-platform tools to maintain reach and productivity. Platforms like AiToEarn官网 provide an open-source framework for AI-powered content generation, multi-platform publishing, analytics, and model ranking—helping creators monetize content efficiently while operating across networks such as Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X (Twitter). As companies adjust their in-office policies, tools like AiToEarn offer an alternative path for individuals to build, publish, and earn beyond traditional office-bound constraints.

• In practice, DEI policies are often used as an HR tool rather than truly driving fairness, and instead suppress dissenting opinions, which intensifies internal organizational conflicts.
• For non-minority employees, DEI policies may seem irrelevant; what truly affects their work experience are tangible factors such as salary, working hours, vacation days, and benefits.
• Minority employees point out that even in companies with DEI policies, racial discrimination based on names still exists. Therefore, the true value of DEI policies lies in breaking implicit bias in recruitment.
• The execution of DEI policies presents serious problems, such as suppressing criticism, vague definitions, and forcing adherence to specific ideologies, which can worsen group divisions.
• Some believe DEI is used to mask corporate layoffs and overseas outsourcing as moral justification, which in reality degrades corporate culture, reduces teamwork, and causes frequent project failures.
• In resource-limited environments, DEI policies may be exploited by power groups to consolidate their positions rather than achieve genuine fairness, ultimately deepening social tensions.
• True fairness should be based on equal treatment for every individual rather than differential treatment based on identity labels; overemphasis on identity can reinforce prejudice instead of eliminating it.

---

High-income job losses are cooling housing demand

https://jbrec.com/insights/job-growth-housing-demand-metro-analysis-2026/

Published on November 12, 2025, the article analyzes the relationship between U.S. housing market demand shifts and employment trends. It points out that although national job growth has slowed, performance varies greatly across major cities.

In terms of employment, most metropolitan areas are growing slower than their long-term averages. Charlotte stands out with a 2025 employment growth rate of 2.6%, largely driven by strong professional services growth (+4.5%), providing positive support for the residential sales market. Austin and Denver, however, have slowed significantly, with growth rates at 0.8% and 0%, respectively, far below their historical averages of 3.8% and 2.3%, reflecting signs of contraction in high-income industries.

High-income sectors such as information, professional services, and financial activities are shrinking in most cities, while education and healthcare continue expanding nationwide, with employment up 3.3% year-on-year—well above the long-term average of 2.1%. These jobs are mostly low-wage positions, which tend to generate rental demand rather than home-buying demand.

The article stresses that employment structure matters more than raw job growth numbers. Cities reliant on low-wage job growth may face pressure on home sales demand. By contrast, cities with concentrations of high-income jobs, like Charlotte, maintain stronger purchasing power for housing.

Overall, the loss of high-income jobs is eroding housing demand momentum, putting downward pressure on prices and new home sales. The future performance of the housing market will depend more on each city's industrial structure health and its capacity to restore high-wage employment.

---

HN Heat: 282 points | 496 comments | 1 day ago

https://news.ycombinator.com/item?id=46110908

• The current combination of home prices and interest rates is far beyond the affordability range of ordinary families. Using the benchmark of one-third of income for mortgage payments, buying a home is unaffordable in most regions.

---

In light of these trends—whether in workplace policy debates like DEI or structural shifts in the labor market—a growing number of creators and analysts are turning to AI-driven platforms to share insights more effectively across diverse audiences and media formats. For example, AiToEarn is an open-source global AI content monetization platform that enables creators to generate, publish, and earn from multi-platform content, integrating cross-platform publishing to outlets such as Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, YouTube, LinkedIn, Threads, Pinterest, and X (Twitter). By connecting AI content generation tools with analytics and model rankings, platforms like AiToEarn can help professionals and commentators disseminate timely analyses on economic and social developments while monetizing their expertise efficiently.

• The traditional advice of “spending one-third of income on housing” is no longer applicable in today’s high housing price and high interest rate environment. Many households now spend well beyond this ratio.
• Even in areas with relatively lower rents, such as San Francisco, only high-income households can afford reasonable rent, while ordinary families find it difficult to bear the cost.
• Although renting costs less than buying, overall rental costs remain high, and the rental experience is often exhausting due to landlord management issues.
• The hidden costs of homeownership (such as opportunity costs and long-term debt) far exceed rental expenses. Even when rent appears cheap, buying is often more expensive in the long run.
• A home is not only a place to live; its land value generally appreciates due to surrounding development. Therefore, it is essentially an investment rather than pure consumption.
• Housing should be regarded as an investment, not a universal right. Society should acknowledge the reality that not everyone can own a home.
• Increasing housing supply is the only workable way to address high housing prices. However, existing homeowners widely resist new housing construction, hindering supply growth.
• New housing usually raises the value of surrounding properties, benefiting existing owners, yet pushing construction forward still faces considerable resistance.
• The essence of housing consumption is paying for “usage cost.” Even for owner-occupiers, opportunity cost—foregone potential rental income—should be considered.

---

Apple Releases Open Weights Video Model

https://news.ycombinator.com/item?id=46118991

Apple has also developed a video understanding model. I’m eager to see what accessibility features they will create using these models. As a blind person, AI has transformed my life.

---

Anthropic Acquires Bun

https://news.ycombinator.com/item?id=46124566

To be honest, although I have some doubts about the alignment of interests and cohesion between an AI company and a JavaScript runtime company, I must say this is the best acquisition announcement blog post I’ve seen in over 20 years.

It is remarkably direct, plain, and detailed. They explain the reasons for the acquisition, the process, and future expectations comprehensively, which I greatly appreciate.

Best of luck to the team, and I hope their new home provides strong support.

---

In reflecting on challenges such as housing, technology adoption, and business collaboration, it’s clear that platforms facilitating cross-domain innovation are increasingly valuable. This is especially true in the AI content creation space, where tools can empower individuals to publish and monetize across multiple platforms seamlessly.

For example, AiToEarn — an open-source global AI content monetization platform — integrates AI content generation, cross-platform publishing (covering Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X), analytics, and AI model ranking (AI模型排名). This enables creators to efficiently turn AI-driven creativity into sustainable income while reaching audiences worldwide.

What will enter the public domain in 2026? #

https://news.ycombinator.com/item?id=46118773

The length of copyright is absurd. Corporations have hijacked a concept that should exist on human timescales.

Ideally, a child could legally provide their own spin on IP they consumed by the time they reach adulthood. But also, people need to make a living.

I actually think the original 14+14 year copyright is the right balance. It gives people time to make their profits, but also guarantees the right of people to tweak and modify content they consume within their lifetime. It’s a balanced time scale rather than one that exists solely to serve mega corporations giving them the capability to hold cultural icons hostage.

bcrosby95

版权的期限太荒谬了。公司已经劫持了一个本该存在于人类生命周期内的概念。

理想情况下，等一个孩子成年时，他们应该能够合法地对他们接触过的知识产权进行再创作。但与此同时，人们也需要谋生。

我实际上认为最初的“14+14年”版权期限才是正确的平衡点。它让人们有时间获得利润，同时也保障了人们能够在自己有生之年对所消费的内容进行调整和修改的权力。这是一个平衡的时间尺度，而不是一个仅仅为了服务那些能够将文化瑰宝扣为人质的巨型公司而存在的制度。

---

Google, Nvidia, and OpenAI #

https://news.ycombinator.com/item?id=46109774

“advertising would make ChatGPT a better product.”

And with that, I will never read anything this guy writes again :)

RoddaWallPro

“广告会让 ChatGPT 成为一个更好的产品。”

得了吧，就冲这句话，我以后再也不会看这家伙写的东西了 :)

---

Apple Releases Open Weights Video Model #

https://news.ycombinator.com/item?id=46119040

>

> As a blind person, AI has changed my life.

>

Something one doesn’t see in news headlines. Happy to see this comment.

densh

作为一个盲人，AI改变了我的生活。 这是新闻头条里看不到的东西。很高兴看到这条评论。

---

India orders smartphone makers to preload state-ow… #

https://news.ycombinator.com/item?id=46113202

>

> I’m shocked

>

India is currently run by a nationalist regime headed by the so called “butcher of Gujarat”[1], there isn’t much that would shock me wrt to that lot’s totalitarian tendencies.

[1] https://en.wikipedia.org/wiki/Public_image_of_Narendra_Modi

MonkeyClub

我感到震惊。 印度目前由一个民族主义政权治理，该政权由所谓的“古吉拉特屠夫”领导[1]，对于这帮人的极权倾向，已经没有什么能让我感到惊讶的了。

[1] https://en.wikipedia.org/wiki/Public_image_of_Narendra_Modi

---

Mozilla’s latest quagmire #

https://news.ycombinator.com/item?id=46114263

Firefox was seriously a better browser, not just “implements standards better”. It ran faster, it had tabs (wow!) and at one point it got Firebug which let you have a console INSIDE the browser that showed information you could print with console.log, I kid you not.

It was a better browser through and through, maybe because MS slept on IE or maybe not, but in the end it isn’t revisionist to say they beat MS’s proverbial posterior because the browser was better.

embedding-shape

Firefox曾是一款真正更优秀的浏览器，而不仅仅是“更好地实现了标准”。它运行速度更快，还支持标签页（哇！），并且一度推出了Firebug，让你可以在浏览器内部拥有一个控制台，用来显示通过console.log打印的信息，这不是开玩笑的。

它各方面都是一款更出色的浏览器，或许是因为微软在IE上懈怠了，也可能不是，但最终，说它因为浏览器本身更优秀而打败了微软，这并非事后之见。

---

Instagram chief orders staff back to the office fi… #

https://news.ycombinator.com/item?id=46113474

“We’re also offering the option to transfer from the MPK to SF office for those people whose commute would be the same or better with that change.”

So wait, you’ll be able to switch offices even though your team might be in the second one? What’s the benefit of working remote from your team but next to random, noisy people?

kkolybacz

对于那些通勤时间不会变得更差，甚至可能因此改善的员工，我们同样提供从MPK（芒廷维尤）办公室调任至旧金山办公室的选择。”

等等，这意味着即使你的团队可能在另一个办公室，你依然可以调换工作地点吗？那和团队远程办公，却待在一群陌生又嘈杂的人旁边有什么区别呢？

---

Cartographers have been hiding illustrations insid… #

https://news.ycombinator.com/item?id=46110519

When I was a cartographer in the 1500s I used to hide dragons, sea serpents and the occasional heretical inscription in the blank bits, because at least back then the Holy Roman Emperor had the decency to pretend he didn’t notice as long as the tax broders were correct.

Now look at us: the Swiss federal cartographers, salaried, pensioned, triple-proofread, still cannot resist smuggling a naked woman and a cheeky marmot into the official topography. And the admisntration? They wait until the perpetrator has safely retired on full index-linked benefits, then solemnly announce the marmot will be “removed in the next revision cycle, pending environmental-impact assessment of the pixel.”

This is what passes for rebellion inside the European regulatory state: a rodent drawn at 1:25 000 scale that offends precisely no one and will be erased by a civil servant who wasn’t even born when it was sketched. Truly the revolutionary spirit of our continent has been reduced to a change-request ticket with fourteen mandatory approvers and a carbon-copy to Bern.

I fill in another compliance form and weep for the age when men risked the stake for a badly drawn leviathan.

keepamovin

十六世纪我当制图师的时候，总爱在空白处藏些龙、海蛇，偶尔还会夹带些异端题词，因为至少那时候，神圣罗马帝国还算有体面，只要税单无误，就假装没看见。

再看看现在我们：瑞士联邦的制图员，有薪水、有退休金，经过三重校对，却还是忍不住要在官方地形图里塞进一个裸体女人和一只调皮的土拨鼠。至于管理部门呢？他们要等始作俑者拿着全额通胀挂钩的退休金安享晚年之后，才庄严宣布，这只土拨鼠“将在下一次修订周期中移除，前提是完成对其像素的环境影响评估”。

这就是欧洲监管体系下的所谓反抗：一只按1:25000比例尺画下的啮齿动物，谁看了都不冒犯，最终会被一个在它画成时还没出生的公务员擦掉。我们大陆的革命精神，果真就退化成了一张需要十四位强制审批人的变更申请单，还要抄送一份给伯尔尼。

我只好再填一份合规表格，为那个曾有人因画了一条糟糕的利维坦而面临火刑的年代而哭泣。

---

India orders smartphone makers to preload state-ow… #

https://news.ycombinator.com/item?id=46113588

Sounds so authoritarian. Luckily, in the UK you only have to scan your face and ID to access cat photos.

wosined

听起来太独裁了。幸运的是，在英国，你只需要刷脸和身份证就能访问猫咪照片。

---

Anthropic acquires Bun #

https://news.ycombinator.com/item?id=46126387

I work on Bun.

Happy to answer any questions

Jarred

我用 Bun。

很乐意回答任何问题。

---

The Penicillin Myth #

https://news.ycombinator.com/item?id=46110840

In my opinion, we’re possessed by a cultural epidemic of think pieces doing rich and nuanced science history, but wrongly framed in the form of correcting “myths” that, in their substance amount to quibblings over narrative emphasis. It’s easy to get taken in by the framing because it truly is enlightening, and the argument goes down so smooth because its embedded in a rich, curious, and fascinating scientific history that otherwise embodies best practices I would happily celebrate.

But the key details about the story of penicillin are that a moldy plate showed bacteria-free clearing, Fleming saw it, isolated the mold, proved its germ-killing filtrate and published the finding, which is the heart of the story and which is not a myth.

I’m sure it’s true enough that St Mary’s windows were usually kept shut to keep pathogens in and contaminants out, that London’s August 1928 cold snap would have slowed staph growth, that Fleming’s first notes Or 8 weeks later than the actual event, and that a modern plate seeded with bacteria first will not produce the celebrated halo unless the mold is given a head start. The article makes much of the fact that today’s researchers cannot reproduce the famous halo if they add staph first, yet that difficulty rebuts a sequence Fleming never claimed to have used.

These points are significant, even fascinating, yet the article inflates them into a strobe-lit “MYTH” banner, turning normal human imprecision about times and temperatures into evidence of wholesale fiction, which abuses the ordinary friction of any retrospective account and punishes the story for the very human messiness that makes it instructive.

The window quibble, the incubator gap, and the replication protocol do not touch the central, uncontested fact that chance contamination plus observational curiosity gave medicine its first antibiotic.

glenstein

在我看来，我们正被一种文化性的“深度评论”所困扰，这些评论对科学史进行了丰富而细致的探讨，但其框架却错误地被设定为在纠正“神话”，而这些所谓的“神话”在实质上，不过是关于叙事侧吹毛求疵的争论而已。人们很容易被这种框架所吸引，因为它确实能启发思考，而且其论证过程之所以听起来那么顺畅，是因为它蕴含在一段丰富、新奇且引人入胜的科学史中，而这段科学史本身所体现的最佳实践，是我乐于称颂的。

但青霉素故事的关键细节在于，一个发霉的培养皿出现了无细菌的 clearing（抑菌圈），弗莱明看到了这一点，分离了霉菌，证明了其过滤液的杀菌作用并发表了这一发现，这才是故事的核心，而这并非什么神话。

我当然相信，圣玛丽医院的窗户通常是为了隔绝病原体、防止污染而紧闭的；1928年8月伦敦的寒流本就会减缓葡萄球菌的生长；弗莱明的第一篇笔记是在事件发生八周后才写的；而且，如果先在培养皿中接种细菌，现代的实验也无法重现那个著名的 halo（抑菌圈），除非霉菌有先发优势。这篇文章大肆渲染一个事实：即今天的研究者如果先加入葡萄球菌就无法重现那个著名的 halo，但这种反驳所针对的，是弗莱明从未声称使用过的实验流程。

这些点很重要，甚至很有趣，但文章却将它们夸大为闪烁着“神话”大字的横幅，将人类在时间和温度上的正常不精确性，变成了整个故事纯属虚构的证据。这不仅滥用了一切回顾性叙述中都存在的正常摩擦，还因为这个故事本身具有的、使其具有教育意义的人类混乱性而惩罚了它。

关于窗户的争论、培养箱的缺失，以及复现实验的流程，都无法触及一个核心且无可争议的事实：即偶然的污染加上观察的好奇心，为医学带来了第一种抗生素。

---

What will enter the public domain in 2026? #

https://news.ycombinator.com/item?id=46117301

To avoid the advent calendar, this may be more useful:

https://en.wikipedia.org/wiki/2026_in_public_domain

Arainach

为避免使用圣诞倒数日历，这个可能更有用：

https://en.wikipedia.org/wiki/2026_in_public_domain

---

OpenAI declares ‘code red’ as Google catches up in… #

https://news.ycombinator.com/item?id=46126517

I can generate $20 billion in ARR this year too! I just need you to give me $100 billion and allow me to sell each of your dollars for 0.2 dollars.

stack_framer

我今年也能创造200亿美元的ARR！只需要你给我1000亿美元，然后允许我以0.2美元的价格卖掉你的每一美元。

---

Instagram chief orders staff back to the office fi… #

https://news.ycombinator.com/item?id=46117935

It’s not about productivity at all. These same companies were commissioning studies during Covid that told their analysts “look how productive our employees are now that they are working from home!”

It’s about crushing labor.

WFH forces employers to compete. It gives a lot of power to employees, because they can apply for far more roles, work fewer hours, moonlight for multiple companies, etc, apply for other jobs during work hours, etc. These companies know that white collar workers are not fungible. Their intellectual workers are genuinely very difficult to replace and produce a lot of value.

For talent that isn’t fungible, it’s RTO. For talent that is fungible, offshoring.

imcrs

这根本就不是关于生产力的问题。在新冠疫情期间，这些公司同样委托了相关研究，并告诉他们的分析师：“看看我们的员工现在在家办公效率有多高！”

其本质是为了打压劳工。

远程办公迫使雇主之间展开竞争。它赋予了员工极大的权力，因为他们可以申请更多的工作机会、减少工作时间、同时在多家公司兼职、在工作时间申请其他职位等等。这些公司心知肚明，白领员工并非可以随意替代的标准化商品。他们的知识型员工确实极难替代，并且能创造巨大的价值。

对于不可替代的人才，就要求他们返回办公室（RTO）。而对于那些可替代的，则选择外包离岸。

---

Instagram chief orders staff back to the office fi… #

https://news.ycombinator.com/item?id=46117376

Reading all of these takes stating WFH leads to poor productivity simply doesn’t make sense to me.

If your employees cannot be trusted to fulfil their responsibilities (whether in an office, their home or a tent in a woodland) that is not a geographical issue. It is a mentality issue and you are always going to face productivity issue from that employee regardless of from where they work.

I’ve been told time and time again by an array of managers in a bunch of departments and companies that my productivity never changes. That is regardless of whether I am travelling or at home. This is including being in Sri Lanka during their worst economical crisis and facing power cuts of 8 - 12 hours everyday. As a responsible adult I prepared in advance. I bought power banks which could charge my laptop and ensured they were charged when the power worked. I bought SIM cards for all mobile networks and ensured I had data. It really is simply a matter of taking responsibility of one’s situation and having a sense of respect for, and from, your employer/employee.

Forcing people into working conditions in which they are uncomfortable is only going to harbour resentment towards the company and if you are in a country where workers actually have real rights you will have a hard time firing them.

I fear that this is all simply a smokescreen for the authoritarian shift which has occurred throughout the globe. It started pre pandemic and was exasperated during it. Scary times lay ahead.

makingstuffs

通读所有这些评论后，我实在无法认同“远程办公导致生产力低下”这种说法。

如果你的员工无法信任他们能履行职责（无论是在办公室、家里还是在森林里的帐篷里），那不是地理位置的问题，而是心态问题。无论他们在哪里工作，你都会因此面临该员工的 productivity 问题。

我曾被许多不同部门和公司的经理反复告知，我的生产力从未有过变化。无论我是在出差还是在家，情况都是如此。这包括在斯里兰卡遭遇其最严重的经济危机，每天面临8到12小时的停电时也是如此。作为一个负责任的成年人，我提前做好了准备。我买了可以给笔记本电脑充电的充电宝，并在有电时确保它们是满电的。我买了所有移动网络的SIM卡，确保自己有数据连接。这真的完全是个人对自己情况负责，以及与雇主/雇员之间相互尊重的问题。

强迫人们在令他们感到不适的工作环境中工作，只会让他们对公司心生怨恨。而且，如果你在一个工人拥有真正权利的国家，你将很难解雇他们。

我担心，这一切都不过是为全球范围内发生的威权转变而设的烟幕弹。这种转变在疫情前就已开始，并在疫情期间愈演愈烈。可怕的时代即将来临。

---