Hacker NewsMonday, May 11

Apple and Google are expanding hardware-based attestation through their APIs to lock out non-approved hardware and operating systems, wrongly presenting this as security while actually eliminating competition. Banks and governments are adopting these systems, with Google's reCAPTCHA now requiring certified smartphones to access web services on desktops. GrapheneOS, despite being more secure, is banned while devices with no security patches for 10 years are permitted, revealing the anti-competitive rather than security-focused nature of these requirements.

Discussion centers on the impressive speed of AI-assisted code translation - Bun's 950k lines converted from Zig to Rust in 6 days. Users debate implications for software development, compute access inequality, and whether this represents a fundamental shift in how code is written and maintained.

Users debate GPT-5.5 Pro's breakthrough performance in solving mathematical problems, noting its superior reasoning and self-correction abilities compared to previous models. However, concerns arise about high costs, token usage, and context limitations that may delay widespread adoption among programmers and researchers.

A former AWS enthusiast describes his 15-year journey from being an early advocate to becoming disillusioned with the platform. Initially excited by cloud computing's revolutionary potential, he gradually became frustrated with issues like expensive egress costs, complex billing, DynamoDB problems, vendor lock-in through Lambda, and AWS's predatory behavior toward open source projects. After moving away from AWS years ago, a recent return for testing triggered a security suspension that disabled his business email for days with poor support response, reinforcing his decision to leave the platform permanently.

Discussion centers on whether 404, 400, or 414 is the correct HTTP response for unexpected query strings, with references to W3C standards showing query strings are part of the URL API. Commenters debate the technical correctness of blocking query parameters.

The content contains two links - one to a page about HTML effectiveness and another to a related Simon Willison article about the unreasonable effectiveness of HTML. However, there is no substantive text content provided to summarize beyond these reference links.

Large Language Models (LLMs) are not ready for delegated workflows where users entrust them with complex document editing tasks. A study using DELEGATE-52, which tests 19 LLMs across 52 professional domains, found that even top models like GPT-4 and Claude corrupt 25% of document content during long workflows. The degradation worsens with larger documents, longer interactions, and distractor files, making current LLMs unreliable for delegation tasks.

Louis Rossmann pledged $10,000 to cover legal fees for developer Pawel Jarczak, who was threatened by Bambu Lab with a cease and desist letter over his "OrcaSlicer-BambuLab" project that restored direct control between Bambu Lab 3D printers and OrcaSlicer software. Rossmann is mobilizing the Right to Repair community to support the developer and crowd-fund his legal defense against what he considers corporate bullying.

Modern software developers are creating fragile, privacy-invading applications by defaulting to cloud-based AI APIs instead of utilizing local device capabilities. The author argues for on-device AI processing, demonstrating how their news app generates summaries locally using Apple's APIs, avoiding privacy concerns, network dependencies, and costs while maintaining functionality and user trust through truly local-first development.

Employees sending unedited AI outputs to colleagues, creating more work for recipients who must parse lengthy, low-quality content. The lack of social norms around AI use leads to asymmetric effort where less capable people generate more content requiring skilled people to spend time processing it.