0:00 / 2:36ai OpenAI ships GPT-5.6 with a cheaper, multi-model push
OpenAI says GPT-5.6 is now generally available after a limited preview, with three models: Sol as the flagship, Terra for everyday work, and Luna as the lower-cost option. The company claims major gains in coding, knowledge work, cybersecurity, science, and cost efficiency, plus an “ultra” setting that coordinates multiple agents in parallel for harder tasks. OpenAI also emphasizes expanded safeguards, including red-teaming, real-time checks, trusted cyber access, and claims that the models do not cross its critical thresholds for biology or cybersecurity capability.
Discussion: Mixed — HN is impressed but not credulous. Many commenters are excited about token efficiency, coding gains, and a possible new alternative to Anthropic’s models, while others worry about cherry-picked benchmarks, confusing product names, quota burn, guardrails, and whether real-world performance will match the launch post. (Strong interest in coding-agent performance versus Claude/Fable, Positive reaction to lower token use and cost-per-task claims, Skepticism about benchmark selection and marketing framing)
0:00 / 1:05science A 2015 snail-tooth strength story crawls back onto HN↻ from 2015
This 2015 Smithsonian piece resurfaced on HN today: it reports that limpet teeth tested by U.K.-based engineers showed tensile strength above spider silk, with the article saying they averaged about five times stronger than most spider silk. The teeth are described as goethite nanofibers in a protein matrix, putting them ahead of Kevlar and around high-quality carbon fibers, though not above materials like graphene. A later editor’s note clarifies an important distinction: the article’s “strength” claim is about tensile strength, not hardness or compressive strength.
Discussion: Mixed — The thread is curious and amused, but much of the energy goes into poking fun at the article’s comparison of tensile strength to spaghetti holding thousands of one-pound sugar bags. A few commenters ask for better images or the original paper, while others riff on slugs biting skin and odd unit conversions. (Interest in the original research and images, Amusement and frustration over confusing weight analogies, Pedantry about units and material-strength terminology)
0:00 / 1:48security EU extends “Chat Control” scanning fight to 2028
The European Parliament allowed the interim “Chat Control 1.0” regime to continue, permitting voluntary scanning of private, unencrypted communications until 2028 or until a permanent regulation replaces it. The linked article says 314 voting MEPs opposed the measure and 276 supported it, but the rejection motion needed an absolute majority of 361 and therefore failed. Former MEP and civil-rights activist Patrick Breyer argues the process undermines democracy and that broad scanning produces false alarms while delaying more targeted child-protection measures. End-to-end encrypted chats such as WhatsApp are described as exempt, while negotiations over a permanent “Chat Control 2.0” law are expected to resume in September.
Discussion: Negative — HN reaction is overwhelmingly hostile, focused on privacy, surveillance, and frustration with EU procedure. Many commenters see the vote mechanics as anti-democratic because a voting majority opposed the regulation but failed to reach the absolute-majority threshold needed to reject it; a smaller set corrects details about which EU body drove the process and notes that end-to-end encrypted chats remain excluded for now. (Suspicionless scanning as mass surveillance, Anger at EU parliamentary procedure and absenteeism, Concern that Chat Control 1.0 paves the way for attacks on end-to-end encryption)
0:00 / 2:10security Apple accuses OpenAI of stealing its hardware playbook
Apple has sued OpenAI, io Products, and former Apple employees Tang Tan and Chang Liu, alleging they stole Apple trade secrets and confidential information for OpenAI’s hardware efforts. The complaint says Tan used internal Apple project knowledge in recruiting, allegedly asked candidates to bring Apple parts, CAD artifacts, and prototypes, and that Liu exploited a security bug after leaving Apple to download confidential engineering files. Apple is seeking injunctions and damages as OpenAI, after acquiring Jony Ive’s io in a $6.5 billion deal, works toward its first consumer hardware device.
Discussion: Negative — HN’s reaction is sharply negative toward OpenAI, with many commenters calling the allegations unusually specific and damaging. A smaller thread urges caution that these are Apple’s claims, not established facts, and separates legitimate employee mobility from taking documents, parts, or confidential details. (Allegations viewed as serious and specific, Strong distrust of OpenAI’s corporate culture and IP practices, Distinction between non-competes and trade-secret theft)
0:00 / 1:55policy John Deere must open repair tools to farmers under FTC settlement
The FTC and attorneys general from Arizona, Illinois, Michigan, Minnesota and Wisconsin reached a settlement requiring Deere & Co. to make diagnostic and repair tools available to equipment owners and independent repair shops, not just authorized dealers. The order also bars Deere dealers from retaliating against owners or shops that do their own repairs, puts Deere under compliance oversight for 10 years, and requires $1 million collectively to the five states for enforcement costs, pending approval by Judge Iain D. Johnston. It follows a separate $99 million class-action settlement with farmers earlier this year and marks a major right-to-repair enforcement action in agricultural equipment.
Discussion: Mixed — HN is broadly pleased to see a right-to-repair win against John Deere, with many treating equipment repair as a basic ownership right. The mood is tempered by skepticism about the small $1 million payment, the 10-year compliance structure, possible loopholes, and debates over emissions-tampering concerns. A large side thread credits right-to-repair activists, especially Louis Rossmann, while arguing over his style and influence. (Right to repair as a basic ownership freedom, Skepticism about enforcement strength and loopholes, Praise for repair-rights activists)
0:00 / 2:35policy NYC takes aim at hard-to-cancel subscriptions and junk fees
New York City has adopted a rule, starting October 1, that bans companies from using deceptive subscription practices that make recurring services hard to cancel. Violators could face $525 per user subscription, back fees, and additional fines if they fail to provide a simple cancellation path. The city is also proposing a separate “junk fee” rule requiring mandatory charges to be included upfront in advertised prices, with potentially large effects on rentals, hotels, and other consumer services.
Discussion: Mixed — HN was broadly supportive of rules against subscription traps and hidden fees, with many commenters sharing frustrations about cancellations, resort fees, restaurant surcharges, and recurring charges. The mood was tempered by skepticism about enforcement, local jurisdiction, possible carve-outs, and whether calling the rule “landmark” overstates it given California’s existing consumer-protection rules. (Strong support for simple cancellation and upfront pricing, Skepticism about whether NYC can enforce the rules effectively, Comparisons with California’s subscription and drip-pricing laws)
0:00 / 1:57general HN falls hard for a tiny timed word game
A Show HN post for 18 Words drew major attention: it appears to be a browser word puzzle where players unscramble 18 words, with commenters focusing on a 30-second timer and daily-challenge style scoring. The discussion became a rapid product-feedback session, with the maker asking about timer-free modes and later claiming an update that lets players continue through all 18 words and share an x/18 result. It matters less as tech news than as a case study in lightweight game design, Wordle-like mechanics, and HN’s ability to stress-test a small project overnight.
Discussion: Mixed — The thread is broadly enthusiastic about the simple, crisp word-game concept, but packed with usability requests. The biggest friction point is the timer and whether failure should end the run; commenters also repeatedly ask for shuffling, alternate languages, better scoring, and accepting multiple valid anagrams. (Praise for a clean, compact puzzle UI, Debate over timed versus relaxed modes, Requests to continue through all 18 words after a miss)