Sixteen bytes, infinite demo art & OpenAI governance and AI race - Hacker News (May 24, 2026)

Let’s start with AI and governance. In a wide-ranging interview on The Knowledge Project, OpenAI co-founder and president Greg Brockman walks through the decisions that shaped OpenAI’s technical roadmap and its shift away from a purely nonprofit setup. But the headline moment is his inside account of the roughly 72 hours after the board fired CEO Sam Altman: the call that kicked it off, Brockman deciding to resign the same day, and the rapid “Phoenix” contingency plan that took shape at Altman’s home. Why this matters isn’t just the drama. It’s a live case study in how fragile governance can be when the stakes are global and the technology is moving fast. Brockman also talks about the broader AI race, argues that progress is accelerating, and claims AI is already writing a meaningful chunk of OpenAI’s software—an indicator that internal productivity gains from AI aren’t theoretical anymore. He also addresses why ChatGPT stopped showing reasoning traces, pointing to safety and product concerns. Whether you agree or not, that decision sits right at the intersection of transparency, misuse risk, and user trust—three issues that are going to define how AI products are judged over the next few years.

Staying in the security lane, there’s a troubling report about scammers abusing Microsoft’s own internal notification system. For months, phishing and spam emails have been sent from an address that typically delivers legitimate account alerts—msonlineservicesteam@microsoftonline.com. That’s the kind of sender users are trained to trust, especially when it comes to sign-in warnings or two-factor codes. The significance here is bigger than one email address. Attackers are increasingly bypassing classic spoofing by piggybacking on authentic delivery channels—systems meant to send real transactional messages at massive scale. If recipients start doubting even “official” security emails, the whole ecosystem gets noisier, and real alerts become easier to ignore. Microsoft says it’s investigating and tightening detection, but the episode is a reminder: trust signals can become liabilities when automation is too flexible.

Now for a piece of computing history that’s both nerdy and genuinely important: Microsoft has open-sourced what it says is the earliest DOS-related source code yet discovered, reaching back to 86-DOS 1.00—before the MS-DOS name even existed. Alongside that are early PC-DOS 1.00 kernel snapshots and classic utilities like CHKDSK, plus notes and documentation. The wild part is how it survived: not as neat digital archives, but as aging paper printouts. A preservation team—nicknamed the “DOS Disassembly Group”—had to scan, transcribe, and reconstruct code that OCR tools couldn’t reliably read. Why it matters: DOS sits at the root of the IBM PC explosion and the entire clone ecosystem that followed. Having primary-source code from that era is invaluable for historians, emulator authors, and anyone trying to understand the technical and business decisions that shaped personal computing.

If you like the idea that constraints can produce creativity, the demoscene delivered a perfect example. A new write-up explains “wake up! 16b,” a 16-byte real-mode DOS program shown at the Outline demoparty earlier this month. It produces a Sierpinski-triangle-like visual pattern in text-mode video memory—while simultaneously generating sound through the PC speaker. What makes it fascinating isn’t just the tiny size. The output depends heavily on whatever data happens to be in memory and on differences between emulators and real hardware. In other words, the computer’s quirks become part of the artwork. This is a great reminder that “software” isn’t always a clean abstraction. At the edges—especially in extreme sizecoding—you end up collaborating with the machine’s history, its analog imperfections, and its implementation details.

On the systems side, ClickHouse has open-sourced something called “silk,” a cooperative fiber scheduler for Linux aimed at running huge numbers of lightweight, stackful coroutines with low overhead. The pitch is straightforward: if you’re building I/O-heavy services, you want concurrency without the cost and complexity of a thread-per-request model. The bigger takeaway is that we’re seeing continued experimentation in how modern servers juggle work—especially as applications lean on asynchronous I/O and try to squeeze more throughput out of the same CPU budget. Even if you never adopt this particular project, it’s a sign of where performance-minded infrastructure teams are spending their time: making high-concurrency code feel simpler without giving up control.

A totally different kind of “performance story” comes from design. Doug MacDowell describes spending roughly 50 hours hand-drawing a statistically accurate line-graph-style visualization—something software could do in minutes. He wasn’t chasing efficiency. He was trying to understand pre-digital drafting methods: the grids, the careful plotting, the inking, and the slow reveal of clean final lines. Why it matters in a tech context: as tooling gets faster and more automated, it’s easy to forget that charts are arguments, not just outputs. The manual approach forces you to make deliberate choices about what you’re emphasizing and what you’re hiding. Even if no one should do this for every dashboard, the exercise is a useful antidote to mindless “export a graph and move on” workflows.

More retro computing—this time from the Commodore 64 repair world. A new post documents the font used by the C64 “Dead Test” diagnostic cartridge, and points out that practical reference material has been surprisingly scarce. The author catalogs what the cartridge actually implements, how it differs from the standard C64 character ROM, and even identifies a strange unused glyph as the MICR “transit” symbol—a neat little historical Easter egg tied to bank-check encoding aesthetics. The technical punchline is the real value: the cartridge can display text without relying on the machine’s built-in ROMs, which is exactly what you want when you’re diagnosing a computer whose ROMs might be broken. Preservation, repairability, and documentation aren’t glamorous—but they’re how old hardware stays usable rather than becoming silent plastic.

To close, a big-picture piece from Quanta Magazine revisits the mathematician Alexander Grothendieck and explains why his work reshaped 20th-century math. His central move was to replace a patchwork of definitions in algebraic geometry with unifying frameworks that focus on relationships and structure—most famously, the idea of a “scheme,” which lets mathematicians reason about solutions to equations in a way that carries across different number systems. If you’re not a mathematician, here’s the “why it matters”: Grothendieck’s abstractions became durable infrastructure. They’re the kind of conceptual tools that make entirely new research programs possible—especially where geometry, algebra, and number theory collide. It’s also a reminder that foundational theory can have a long half-life: decades later, people are still building on the highway he laid down.