Top Stories

Git Commands I Run Before Reading Any Code

1846 points · piechowski.io

The runaway hit of the day is a deceptively simple idea: before opening a single source file, run five git commands that reveal where the codebase hurts. The author walks through identifying churn hotspots (files changed most often), contributor analysis (who built what and where’s the bus factor risk), bug clustering (which files keep breaking), commit velocity (is the project gaining or losing momentum), and firefighting frequency (how many reverts and hotfixes signal deploy fear). The 391-comment thread is a goldmine of engineers sharing their own diagnostic rituals. The appeal is obvious — in minutes you get a diagnostic picture of a project’s health that would take days of code reading to piece together.

I Ported Mac OS X to the Nintendo Wii

1307 points · bryankeller.github.io

Bryan Keller pulled off what most would dismiss as impossible: running Mac OS X 10.0 (Cheetah) on a Nintendo Wii. The Wii’s PowerPC 750CL processor and 88MB of RAM turned out to be just barely enough, but the real challenge was everything else — writing a custom bootloader from scratch, implementing drivers for the Wii’s Hollywood system-on-chip, wrangling a reversed-little-endian USB stack, and patching the kernel for the Wii’s unusual memory layout. The result is a fully functional Mac OS X desktop on a game console. The 218-comment thread is equal parts technical admiration and nostalgia for an era when Apple shipped on PowerPC. A magnificent feat of low-level engineering.

VeraCrypt Crisis: Microsoft Terminates Account Without Warning

1157 points · sourceforge.net

The biggest open-source drama of the day: Microsoft abruptly terminated the account VeraCrypt’s lead developer Mounir Idrassi uses to sign Windows drivers and the bootloader — with no prior warning or explanation. Since Windows represents the vast majority of VeraCrypt’s user base, this effectively halts new releases for the platform that needs them most. The 427-comment thread on SourceForge (plus 185 more on a related 404 Media story at 483 points) is a mix of outrage and concern about the fragility of open-source supply chains that depend on platform gatekeepers. Microsoft VP Scott Hanselman stepped in publicly to help, and other projects like WireGuard reported similar terminations, suggesting a systemic issue rather than targeted action. The situation appears to be moving toward resolution, but the incident is a stark reminder of how much critical security infrastructure depends on the goodwill of platform vendors.

US Cities Are Axing Flock Safety Surveillance Technology

646 points · cnet.com

A growing number of US cities are dismantling Flock Safety’s automated license plate reader networks amid mounting privacy concerns. Flock’s cameras, which can track vehicle movements across entire metro areas, have been controversial since their rapid deployment — often installed with minimal public input or oversight. The 382-comment thread reflects a broad HN consensus that the surveillance-to-safety tradeoff isn’t worth it, with many pointing out that the data retention policies and law enforcement access provisions lack adequate safeguards. The pushback marks a notable shift: cities that enthusiastically adopted the tech are now actively removing it.

ML Promises to Be Profoundly Weird

406 points · aphyr.com

Kyle Kingsbury (of Jepsen distributed systems testing fame) delivers a characteristically sharp essay on the fundamental strangeness of large language models. His core argument: LLMs are simultaneously capable and incompetent in ways that make their actual utility maddeningly hard to assess. They handle advanced mathematics but fail at simple logic puzzles. They generate convincing prose but stumble on basic spatial reasoning. The 436-comment discussion is one of the most substantive AI debates HN has had in months, with practitioners sharing their own experiences of the “jagged competence frontier” — the idea that LLM capabilities are wildly inconsistent across domains in ways that resist easy characterization.

Meta Introduces Muse Spark: Scaling Toward Personal Superintelligence

287 points · ai.meta.com

Meta Superintelligence Labs shipped its first product: Muse Spark, a natively multimodal reasoning model with tool use, visual chain-of-thought, and multi-agent orchestration. The headline feature is “Contemplating mode,” which orchestrates multiple parallel-reasoning agents to tackle complex problems — achieving 58% on Humanity’s Last Exam and 38% on FrontierScience Research. The model also offers interactive health explanations developed with over 1,000 physicians. The 301-comment thread is predictably split between those impressed by the benchmarks and those questioning whether Meta’s “personal superintelligence” framing is aspirational marketing or a meaningful roadmap. Available now at meta.ai and through a private API preview.

MegaTrain: Training 100B+ Parameter LLMs on a Single GPU

264 points · arxiv.org

A new paper demonstrates training models up to 120 billion parameters on a single H200 GPU at full precision — no model parallelism required. MegaTrain’s trick is inverting the traditional GPU-centric approach: it treats the GPU purely as a compute engine while storing model parameters and optimizer states in host CPU memory, then uses pipelined double-buffered execution to keep the GPU fed continuously. The results are striking: 1.84x the throughput of DeepSpeed ZeRO-3 with CPU offloading for 14B models, and 7B model training with 512k token context on a GH200. The 49-comment thread is excited about the democratization angle — making large-scale training accessible to researchers without multi-GPU clusters.

Who Is Satoshi Nakamoto? The Quest Continues

331 points · nytimes.com

The New York Times publishes a deep investigation into the identity of Bitcoin’s creator, with the evidence trail pointing toward Adam Back, the inventor of Hashcash (whose proof-of-work system directly inspired Bitcoin’s mining mechanism). The 269-comment thread is doing what HN does best with Satoshi stories: meticulously picking apart the evidence, debating whether unmasking is ethical, and arguing over whether it even matters at this point. The piece arrives at an interesting moment, with Bitcoin’s institutional adoption now firmly established and the philosophical question of anonymous creation more relevant than ever.