Published on May 24, 2026 at 06:00 CEST (UTC+2)
Microsoft open-sources "the earliest DOS source code discovered to date" (87 points by DamnInteresting)
Microsoft open-sources "the earliest DOS source code discovered to date"
Microsoft has released what it claims is the earliest known DOS source code, including 86-DOS 1.00 kernel snapshots and utilities like CHKDSK, predating the MS-DOS branding. The article provides historical context: programmer Tim Paterson created 86-DOS for Seattle Computer Products, and Microsoft licensed and later bought it to supply an operating system for the IBM PC. This release continues Microsoft’s pattern of periodically open-sourcing foundational code, offering a rare glimpse into the origins of PC operating systems.
Scammers are abusing an internal Microsoft account to send spam links (65 points by spike021)
Scammers are abusing an internal Microsoft account to send spam links
Scammers have been exploiting a loophole in Microsoft’s account system to send spammy emails from the official msonlineservicesteam@microsoftonline.com address—the same one used for legitimate security alerts like two-factor codes. The emails mimic official fraud warnings or private message notifications, potentially tricking recipients. Microsoft has not yet resolved the issue, highlighting a significant security gap in identity verification and internal mail flow control.
Wake up! 16b (89 points by MaximilianEmel)
Wake up! 16b
This is a write-up about a 16-byte x86 assembly demo released at the Outline Demoparty, showcasing extreme algorithmic density. The author explains how they achieved a self-contained program that generates graphics and sound using techniques like polymorphic instructions and jumping into the middle of opcodes. The piece reflects on decades of sizecoding experimentation and the surprising depth of behavior that can emerge from just a handful of bytes.
Time to talk about my writerdeck (322 points by hggh)
Time to talk about my writerdeck
The author describes converting an old Linux laptop into a distraction-free writerdeck by installing only a console (tty) version of Debian—no desktop environment. The setup includes tools like tmux, neovim, Syncthing for backup, and a few utilities for Wi-Fi and battery management. The goal is to create a dedicated writing device that avoids modern internet distractions while remaining portable and comfortable for long typing sessions.
Green card seekers must leave U.S. to apply, Trump administration says (671 points by tlhunter)
Green card seekers must leave U.S. to apply, Trump administration says
This New York Times article (content not available) reports a new policy requiring green card applicants to leave the United States during the application process. The rule is part of the Trump administration’s broader immigration restrictions, likely affecting many visa holders who previously could adjust status from within the country. The policy shift has significant humanitarian and legal implications, drawing widespread criticism.
On The (2021)
(367 points by ravenical)
On The (2021)
The article advocates for the underused HTML <dl> (description list) element, explaining its proper usage to represent name-value pairs. It details the syntax of <dl>, <dt>, and <dd> and gives real-world examples like metadata tables, glossaries, and rent breakdowns. The piece argues that many common UI patterns could be more semantically correct and accessible by using <dl> instead of generic <div> or <table> structures.
My two-part desk setup (2025) (238 points by James72689)
My two-part desk setup (2025)
The author describes reorienting their desk from facing a wall to facing the room, inspired by museum displays. They then split the desk into two halves: a digital side (monitor, keyboard, etc.) and an analog side (paper, notebooks, sketching). The change aims to reduce the pull of digital tools and encourage more diverse creative activities. The article reflects on spatial psychology and the impact of physical environment on work habits.
My I3-Emacs Integration (40 points by nosolace)
My I3-Emacs Integration
The author explains how they patched the i3 window manager to create seamless keyboard binding integration with Emacs, overcoming latency issues from earlier script-based approaches (using xdotool). They maintain common keybindings between i3 and Emacs while still supporting normal graphical windows (e.g., Steam games) that don’t work well under EXWM. The project demonstrates deep customization for users who want the flexibility of a tiling WM with the editing power of Emacs.
Judson's Last Ride (61 points by NaOH)
Judson's Last Ride
This RealClearPolitics article (content not available) appears to be a political commentary piece. The title suggests a narrative about a person named Judson, possibly related to political events or obituary-style coverage. Without access to the content, no further details can be provided beyond the headline’s political context.
Sales and Dungeons: Thermal printer TTRPG utility (62 points by hyperific)
Sales and Dungeons: Thermal printer TTRPG utility
This tool lets users turn a thermal printer into a tabletop RPG companion, printing custom handouts, spells, character notes, or random generators (names, monsters, dungeons). It supports LLM integration for generating content and translations, and includes a session grid that can be shared over a local network so players can trigger prints from their phones. The utility runs on Windows, macOS, Linux, and Raspberry Pi.
LLM-powered creativity tools are entering niche hobbyist markets
The Sales & Dungeons app explicitly integrates LLMs (via OpenAI/OpenRouter) for generating RPG content, translating text, and producing more organic random results. This shows a growing trend of embedding AI not just in productivity or coding tools, but in small-scale, creative utilities for tabletop gaming. Why it matters: It democratizes generative AI, allowing non-developers to leverage large language models for highly specific, low-stakes creativity tasks. Implication: We can expect more “AI in a box” applications where the model is a pluggable component in otherwise deterministic software, increasing the surface area for ethical considerations (e.g., copyright of generated content, bias in fantasy names).
Algorithmic density and minimalism challenge the notion of AI as the only path to intelligence
The “wake up! 16b” demo achieves complex graphics and sound in just 16 bytes of x86 assembly through hand-crafted tricks like polymorphic instructions and jumping into instruction midpoints. This extreme sizecoding is a counterpoint to the heavy reliance on large neural networks – it demonstrates that enormous behavior can emerge from extremely compact, algorithmically dense code. Why it matters: It reminds the AI/ML community that symbolic, hand-optimized logic can be vastly more efficient than learned models for certain tasks, and that “intelligence” (or emergent behavior) can be designed without training data. Implication: For embedded or edge AI, there is renewed value in combining tiny algorithmic kernels with learned components rather than relying solely on deep learning.
Security vulnerabilities in large-scale AI/ML infrastructure are a growing concern
Article 2 describes scammers abusing an internal Microsoft email account to send spam – a classic attack that could also be leveraged against AI/ML systems (e.g., using a trusted service account to distribute phishing links or poison training data). With many AI services relying on email-based authentication and notifications, the ability to spoof or compromise such accounts becomes a critical risk. Why it matters: AI/ML pipelines often have complex chains of integrations (email alerts, API keys, automatic retraining triggers) that present new attack surfaces. Implication: Organizations need to implement zero-trust principles for internal system accounts and consider AI-specific security measures (e.g., anomaly detection for account behavior, cryptographic signing of communications).
Distraction-free computing and “analog” workflows gain traction as AI assistance becomes pervasive
The writerdeck and two-part desk setup articles both advocate for reducing digital overload and creating dedicated spaces away from notifications, web browsers, and AI-driven suggestions. This is a counter-movement to the increasing embedding of AI everywhere – users actively seek constraints that limit automated assistance. Why it matters: As AI becomes ubiquitous, a segment of users will prioritize slower, more deliberate, human-led creation. This is analogous to the rise of film photography or vinyl records. Implication: AI/ML product design should offer explicit “off” modes or opt-in-only assistance, and user research should explore when AI enhances vs. degrades the creative flow.
Semantic HTML and accessibility are undervalued in AI-generated content pipelines
The
Customization and integration of AI tools with existing workflows (e.g., Emacs, i3) remains a friction point
Article 8 demonstrates a developer patching a window manager to achieve low-latency integration between Emacs and i3 – a classic example of the “build your own IDE” mentality. As AI copilots and assistants become more common (e.g., GitHub Copilot, ChatGPT plugins), the ability to deeply integrate them into highly customized environments is often poor. Why it matters: Power users who rely on bespoke setups (tiling WMs, modal editors) may be left behind by AI tools that assume a standard GUI. Implication: AI/ML tools should expose composable, keyboard-driven interfaces and allow users to bind AI actions to custom key combinations or shell commands.
Open-sourcing historical software foundations enables AI/ML to learn from earlier design philosophies
Microsoft’s release of the earliest DOS source code provides a unique dataset for AI systems studying software evolution, operating system design, and low-level optimization. With AI increasingly used for code generation and reverse engineering, access to historically significant, well-documented codebases can improve training data quality. Why it matters: The ability to analyze how resource constraints (e.g., 8086 assembly) shaped design decisions can inform more efficient neural architectures. Implication: AI/ML researchers should actively archive and annotate historical source code – not just as nostalgia, but as a resource for training models that understand trade-offs between performance, size, and clarity.
Analysis generated by deepseek-reasoner