10.13.26 | Hackers trigger European fridge defrosting, Modern Web App in C, GLM-Legal

SPEAKER_1 (0:00)

Welcome to the Hacker News Highlights, where we explore the top 10 posts on Hacker News every day. Today we dive into hackers activating supermarket fridges' automatic defrosting across Europe, building modern web applications in C using CGI, and exploring GLM legal. Let's get started.

Title, Hackers Activate Automatic Defrosting in Supermarket Fridges Across Europe.

Source, Tom's Hardware. The news story reported that cyberattacks disrupted refrigeration systems in French supermarket chains, including Intermarché, Leclerc, and Cooperative U. Attackers apparently forced Danfoss-controlled units into repeated defrost cycles at 50 degrees Celsius, damaging refrigerated and frozen food. The attacks also affected deliveries and electronic shelf labels, and retailers began investigations and repairs. In the comments, the sentiment was mostly alarmed and critical of insecure industrial systems. Users debated whether commercial refrigeration needed remote control at all, or whether it should provide monitoring through isolated networks while keeping control local. Supporting points included reports that similar attacks had affected United States military commissaries, and that Danfoss controllers had exposed authentication flaws and outdated firmware. Overall, users concluded that remote monitoring had value, but Internet accessible control created an avoidable risk.

Title, A Modern Web App in CSource Profit.me The article described a commercial web product built with C, OpenBSD, OpenBSD's HHTP, SQLite, and CGI. The application used one statically linked binary, embedded its assets, avoided JavaScript, and relied on a small set of libraries for payments, email, security, and JSON. The author praised CGI for its simple deployment, clear debugging model, small dependency set, and strong sandboxing through OpenBSD tools. The article also acknowledged that C required hand-built utilities for cookies, forms, validation, escaping, and other common web tasks. In the comments, the sentiment was divided between enthusiasm for the stack's simplicity and concern about C's safety and poor text handling. Users debated whether CGI's process-per-request model was practical, or whether fast CGI, Go, Rust, or a custom server would provide better performance and memory safety. Supporters praised SQLite, minimal dependencies, mature tools, and the ability to understand the whole system. Critics pointed to memory bugs, missing observability, concurrency limits, unclear source availability, and the website's difficult visual design. Overall, users treated the project as a valid small scale experiment, not a general replacement for modern web stacks.

Title, GLM Legal, Source, z.ai. The post appeared to announce GLM Legal, a legal-focused artificial intelligence model from z.ai. The story content was unavailable, but the discussion focused on legal research, drafting, benchmark performance, and human review. Commenters also questioned whether the service was open, which jurisdictions it supported, and whether confidential legal data should be sent to its servers. In the comments, the sentiment was mostly skeptical, with users warning against trusting an artificial intelligence system with legal decisions. Users debated whether lawyers should disclose artificial intelligence use and remain liable for errors, or whether existing professional rules already covered that responsibility. Concern centered on hallucinated cases, weak benchmark results, privacy risks, and the difficulty of detecting missing arguments rather than merely checking cited sources. Overall, users saw GLM Legal as useful for routine work, but unsafe without experienced human review.

Title, AI is a tool and using it well takes skill. Source, wake.org. The article responds to Rich Hickey's claim that large language models are automation rather than tools. It argues that people can wield LLMs through careful prompts, feedback, testing, and correction. The article compares LLMs with compilers and spreadsheets, which automate work but still require human direction. It warns that outsourcing thinking is a misuse of AI, not proof that every use of AI abandons human judgment. In the comments, the sentiment was divided between support for AI as a powerful tool and concern that it encouraged cognitive decline. Users debated whether unpredictable outputs could count as tool behavior, since traditional tools offered clearer feedback and more reliable results. Supporters said effective use required domain knowledge, clear communication, verification, context management, and guardrails. Overall, commenters saw AI as useful but dangerous when it replaced understanding, instead of extending it.

Title, Turn a WiFi 7 Router into a 320 MSP Software-Defined Radio Receiver. Source, GitHub. The project turns a Gemtek W1700K WiFi 7 Router into a wideband software-defined radio receiver. Its MediaTek radio captures 14-bit complex samples at up to 320 MSP across the 2.4, 5, and 6 GHz bands. Custom firmware streams the data over 10 Gbit Ethernet while Linux tools provide a live spectrum display, tuning, recording, and gain controls. The receiver is receive-only and replaces the router's original software. In the comments, the sentiment was largely positive, with users calling the project a major advance for affordable radio work. Users debated whether the project showed genuine understanding, or mostly demonstrated what an artificial intelligence coding tool could achieve. Supporters said the hardware delivered unusually high bandwidth and sample quality for a low-cost device, while critics wanted clearer documentation and more evidence about calibration, dynamic range, and the firmware's inner workings. Overall, users saw it as a valuable starting point for open radio experimentation.