Hacker News
1Hang on to Your Firefox
- A blogger defends Firefox against criticism, arguing that people shouldn't abandon it despite minor annoyances like its presence on social media platform X. The author emphasizes that Firefox is crucial for browser engine diversity and competition, serving as the last independent alternative to Google Chrome's dominance, and urges users to support it given its declining market share.
Firefox criticism comes from users who care
Power users criticize Firefox not out of hatred but because they care deeply about browser diversity. They're frustrated by Mozilla's questionable decisions like ad-tech acquisitions and AI features, but continue using Firefox as the last alternative to Chrome's dominance.Firefox vs Chrome technical performance
Users report mixed experiences with Firefox performance, including RAM usage issues and mobile interface problems. Some praise Firefox's dev tools and features like tree-style tabs, while others find Chrome more reliable for certain tasks.Firefox's superior ad blocking capabilities
Firefox remains the only major browser supporting full uBlock Origin functionality, while Chrome's Manifest V3 limits ad blocking effectiveness. This advantage is seen as Firefox's strongest selling point for privacy-conscious users.
2How accurate have Ed Zitron's AI skeptic predictions been?
- The author analyzed AI skeptic Ed Zitron's predictions and found them consistently wrong. Zitron claimed major tech companies like Meta, Google, and Microsoft were "dying" and desperately turning to AI, but these companies actually showed strong revenue and profit growth. The author argues Zitron uses misleading data and flawed reasoning while positioning himself as an authority who "looked at the numbers," when his analysis doesn't support his conclusions.
Ed Zitron's prediction accuracy
Commenters debate whether Zitron's AI predictions should be judged on timing vs. underlying assessment. Some argue his financial analysis is correct despite wrong timelines, while others note subsidized usage inflates current AI metrics.AI industry financial sustainability
Discussion centers on whether massive AI investments are justified by actual demand. Users point to subsidized usage, circular financing between companies, and questions about whether revenue growth can support infrastructure costs.Evaluation methodology and bias
Debate over whether people are reinterpreting Zitron's failed predictions to fit their own views rather than evaluating his actual claims. Some defend focusing on underlying analysis over specific predictions.
3Gemini 3.8 Flash and 3.8 Flash Cyber
Performance and benchmarks
Gemini 3.8 Flash achieves top performance on DeepSWE and matches Opus 5 medium with intelligence score of 59. Users debate benchmark reliability vs real-world performance, with some skeptical of "benchmaxxing" while others praise the model's speed and cost-effectiveness at 1/5th the price of competitors.HTML/JavaScript generation capabilities
Multiple users tested Gemini's ability to create "cool HTML things," generating particle simulations and interactive demos. Notable for speed (13 seconds, 1.8 cents) but some outputs had issues like hardcoded "60 FPS" displays. Users compared results across different models and discussed the prevalence of particle simulations in LLM outputs.Real-world applications and multimodal features
Users highlight Gemini's strengths in trip planning, real-world knowledge, photo analysis, and document parsing. The model's multimodal support for audio/video input gives it advantages over image-only competitors like Claude and GPT models, making it valuable for media analysis tasks.
4A note on subscription prices from LWN
- LWN is raising subscription prices by approximately 20% starting September 15, the first increase in nearly five years. The online publication cites inflation and rising costs as reasons for the change. Monthly prices will range from $6 for starving hackers to $55 for supporters.
High quality and value of LWN
Commenters consistently praise LWN as one of the highest-signal tech publications, crediting their user-funded model for maintaining quality without advertiser influence. Many describe it as foundational to their careers.Pricing and subscription support
Users express willingness to pay increased subscription fees, with many planning to renew or donate more given LWN's value. Some suggest the pricing could be higher given the quality provided.Misleading article title concerns
Multiple readers were alarmed by the vague title, initially fearing LWN was shutting down. LWN editors acknowledged this and updated the title to be clearer about it being a pricing announcement.
5Introducing Ad Blocker for Firefox on iOS
- Firefox for iOS now includes a built-in Ad Blocker that uses Apple's WebKit technology and EasyList filters to block third-party ads and trackers. The feature is optional and can be enabled in Settings, helping reduce screen clutter while browsing. It won't block all ads, particularly those served directly by websites or in search results, and works alongside Firefox's existing privacy protections.
Limited rollout frustration
Users complain about Firefox's gradual experimental rollout of ad blocking, with many still waiting weeks for the feature to appear despite marketing announcements. Some report having access temporarily before it disappeared.Google search ads exemption
The ad blocker doesn't block Google search ads or YouTube ads, likely due to Mozilla's financial dependence on Google. Users share stories of scam ads in search results causing real financial harm to victims.Comparison with existing solutions
Discussion centers on whether Firefox's new ad blocker offers advantages over established alternatives like uBlock Origin Lite on Safari, with most users finding existing solutions superior.
6The ChatGPT/Codex app bundles a full copy of LibreOffice
- A user discovered that the OpenAI Codex desktop app (now ChatGPT) stores 1.7GB of data in their cache folder, including complete Python and Node.js installations, plus native binaries for tools like Poppler, git, and LibreOffice, with skills that teach the app how to use these programs.
Bundling dependencies and bloat concerns
Users discuss whether bundling LibreOffice is excessive bloat or necessary functionality. Some compare it to JavaScript frameworks pulling large dependencies for simple tasks, while others defend it as practical for reliable document handling.OpenAI's development practices and product quality
Critics argue the ChatGPT app is poorly designed, with confusing organization and performance issues. Some suggest it's "vibe coded" and question OpenAI's software engineering quality compared to their AI capabilities.Competition with Microsoft Office and licensing issues
Discussion centers on AI potentially making Office obsolete as a creation tool, turning it into just a viewer. Also covers potential MPL 2.0 license violations from bundling LibreOffice without proper attribution.
7Restroom Archive
- This appears to be a chronological list of bathroom reviews or documentation from various locations including restaurants, cafes, museums, airports, and other establishments across multiple cities like Brooklyn, Manhattan, Salt Lake City, and international locations in Spain and France, spanning from 2023 to 2026.
Stress-inducing countdown timers
Discussion about automatic public toilets with countdown timers that open doors after a set time limit. Users debate whether this feature is stressful and its prevalence in Europe, with mixed experiences reported.Project creator engagement
The creator of the restroom archive project engages with the community, discussing design decisions, accessibility considerations, and inspirations. Users praise the project and suggest contributions and connections to other mapping initiatives.Public bathroom philosophy
Broader discussion about public restroom access, paid vs free facilities, and the tension between privatization and public commons. Users debate business models and the social implications of bathroom accessibility.
8FBI Probes Service Selling 153M+ Drivers Licenses
- A new dark web service called Nexus is selling digital scans of over 153 million drivers licenses from the US and Canada. The breach appears linked to idscan.net, a Louisiana-based identity verification company used by major clients including Hertz, Target, and marijuana dispensaries. The FBI has opened an investigation.
Government-issued digital identity systems
Discussion of how other countries like Estonia, Spain, and European nations have implemented government-issued digital ID systems with cryptographic keys, contrasting with the US's reliance on private verification companies and missed opportunities with REAL ID implementation.Data retention and liability
Debate over why companies retain massive amounts of identity verification data instead of deleting it after verification, with calls for strict liability laws, statutory damages per violation, and treating personal data like radioactive material requiring careful handling.Security concerns and breaches
Discussion of how identity verification services collect detailed personal data including facial scans and government IDs that create valuable targets for attackers, with 153 million records representing roughly half of eligible Americans being compromised.
9Can I opt out of my input or output data being used for training?
- Mistral may use your input and output data for model training in certain cases. You can opt out through the Admin panel for Vibe services (web and mobile) and Mistral Studio/API services. The opt-out process varies depending on which platform you're using.
Mistral's training data policy changes
European organization disappointed with Mistral's shift to opt-in training by default on Team tier, removing admin controls. Users report confusion over conflicting documentation and terminology around "opt-in" vs "opt-out" settings.Trust and verification of AI providers
Users debate whether AI companies actually honor privacy promises, with some arguing all providers train on user data regardless of settings. Discussion includes legal protections, enterprise contracts, and inherent skepticism about vendor claims.Privacy-focused alternatives and local models
Users discuss Duck.ai as privacy-focused alternative and recommend running local models for true data protection. Conversation covers sovereignty concerns and performance comparisons with other AI providers.
10Muse Spark 1.3
Pelican bicycle SVG generation
Users tested Meta's Muse Spark 1.3 model by generating SVG images of pelicans riding bicycles. The new version showed improvements over 1.2, with better bicycle frames and pelican details. Discussion explored why AI models consistently produce similar compositions (side-view, left-to-right movement) due to training data patterns and regression to mean representations.Contributor pricing model
Meta introduced transparent pricing tiers: regular API access vs heavily discounted "contributor" rates (20x cheaper at $0.10-0.20/Mtok) for users who allow training on their data. Users praised the transparency while debating potential security risks of feeding secrets into models and questioning the training value of repetitive tasks.Coding performance and capabilities
Muse Spark 1.3 achieved top DeepSWE benchmark scores (75.4), outperforming competitors while being extremely cost-effective. Users appreciated earlier versions for being tool-like without unwanted "helpfulness," though some worried newer versions might become overly opinionated as Meta pursues better benchmarks.