Hacker News
1.name Termination
- Verisign and ICANN approved the termination of all 3rd-level .name domains in February 2026, affecting 22,000 users including the author who has used neil.fraser.name for 25 years. This will destroy his website, email, and IoT services, while potentially allowing others to hijack his identity by registering the freed fraser.name domain.
Termination of existing domains
Users criticize Verisign's decision to terminate all existing 3rd-level .name domains, arguing they should honor existing registrations and continue reserving 2nd-level domains. Many see this as breaking promises to customers who paid for multi-year registrations.Cookie security and technical issues
Discussion about how .name's structure creates security problems where different users can set cookies across shared 2nd-level domains, and how the Public Suffix List doesn't properly handle .name's unique registration model.ICANN oversight and corporate accountability
Users question ICANN's role in allowing this change despite their mission of ensuring stable internet operations, with some suggesting regulatory intervention and criticism of privatizing internet governance functions.
2GPT-6 Astra
- OpenAI has released GPT-6 Astra with a system card detailing its capabilities. The model shows significant improvements on ARC-AGI-3 benchmarks and demonstrates major gains in coding agent performance according to recent evaluations and ongoing discussions.
Benchmark reliability and AGI definition
Debate over whether high ARC-AGI-3 scores indicate true AGI, with concerns about misleading harnesses inflating performance. Many argue current benchmarks don't capture general intelligence and that AGI requires broader capabilities than test performance.Model capabilities and comparisons
Mixed reactions to GPT-6 Astra's performance versus competitors like Fable 5.1, with users reporting varied experiences. Some praise Astra's improvements while others question independent benchmark results showing it trailing Anthropic models.Rapid development pace fatigue
Users express exhaustion with constant model releases, pricing changes, and difficulty making informed choices. The relentless pace makes it hard to properly evaluate models before they're superseded by newer versions.
3Audacity 4.0
- Audacity 4.0 rebuilds the application interface on Qt and introduces a new clip-editing model with direct clip selection, grouping, and flexible placement between tracks. The interface features movable toolbars, workspace layouts, and new themes. Tool modes are now context-sensitive, and the playhead remains visible during navigation.
Tantacrul's video and background
Users highly recommend Tantacrul's video explaining Audacity 4.0 development. Discussion covers his journey from YouTube critic of music software UIs to Head of Software at Muse Group, his other videos on music notation and software critiques, and recent departure from the company.Technical audio issues and JACK support
Users report persistent problems with Audacity's JACK implementation, including non-persistent clients, poor naming conventions, and broken Linux audio integration. Some express frustration that UI improvements were prioritized over core audio functionality fixes.UI improvements and Qt6 migration
Users praise the new Qt6-based interface as "delicious" and well-designed, noting significant usability improvements beyond cosmetic changes. Discussion includes comparisons to the old wxWidgets interface and appreciation for the concise release video presentation.
4Three sites made 215,128 “best software” pages for AI. Perplexity cites them
- Researchers tested web-grounded AI models by asking for software recommendations across 380 categories. They found 59.8% of citations pointed to low-ranking domains, with 23.4% not in the top million websites. Three related sites published over 215,000 machine-generated "best software" pages since December 2023, explicitly targeting AI retrieval systems with titles like "Facts & Grounding Page" rather than human readers.
LLM self-preference bias
LLMs consistently favor their own generated content over human-written text, even when the human version is objectively better. This bias extends to web search results, code snippets, and content evaluation, creating circular feedback loops.Perplexity quality decline
Users report significant degradation in Perplexity's search quality over time, with faster but less accurate results, mismatched citations, broken features, and poor customer service despite heavy marketing investment.AI-generated web pollution
The internet is becoming flooded with AI-generated content that LLMs then train on, creating a feedback loop of increasingly unreliable information and the potential collapse of trustworthy online knowledge sources.
5Google avoids a breakup of its ad tech business
Merger and breakup asymmetry
Users debate whether mergers and company splits should have equal regulatory barriers. Some argue it's too easy to merge but nearly impossible to undo anti-competitive damage, while others note companies do spin off divisions regularly for business reasons.Google's ad tech revenue definition
Confusion over Google's "ad tech" classification, which refers specifically to third-party ad exchange business (AdX) worth $30B, not Google's main advertising revenue from search/YouTube. Users struggle to understand the distinction between ads and ad infrastructure.Antitrust enforcement effectiveness
Discussion on why tech giants consistently avoid meaningful breakups despite monopoly findings. Users cite companies now "pre-gaming" regulations with former regulators on staff, creating plausible deniability and avoiding clear violations.
6Any Human Ever – One life, drawn at random from all who have ever lived
- This appears to be an interactive experience or game where users can explore human history by randomly selecting from over 100 billion people who have ever lived, then drawing details about a specific year, place, and life from real historical data.
Inaccurate data and AI hallucinations
Users discovered contradictory statistics (96% married vs 44% died before age 15) and traced citations to LLM-generated content rather than real sources. Many criticized the site for spreading misinformation disguised as historical facts.Incorrect probability distribution
Statistical analysis revealed the site doesn't properly sample from birth probability distributions - it overrepresents ancient periods instead of modern times where most humans were actually born, contradicting the "random person" premise.Appreciation for modern life
Despite data accuracy issues, users found the experience thought-provoking and emotionally impactful, with many expressing gratitude for modern healthcare, longevity, and living conditions compared to historical hardships.
7Qwen 3.8 27B available on Cerebras at 1500 tokens/s
- Cerebras offers various open-source models on public endpoints with free trial and pay-as-you-go options. All models are original, unpruned versions using selective weight quantization for storage while maintaining full precision during operations. Pruned research models are available separately on Hugging Face but not through the production API.
Rate limiting issues
Users report that the 150k TPM limit makes Cerebras unusable for coding tasks, with some hitting limits in 90 seconds. At 1500 TPS, users can only run for brief periods before waiting for cooldowns, making sessions only marginally faster than alternatives.No prompt caching discount
Cerebras charges full price for cached tokens unlike other providers, making it 5.6x more expensive than alternatives. This pricing model makes longer sessions prohibitively costly, with users burning through money rapidly on multi-turn conversations.Model performance mixed
Users have conflicting experiences with Qwen 3.8 27B - some find it exceptional for coding while others report it spinning in circles and losing context. The 128k context limit is also seen as restrictive for larger coding tasks.
8Pre-Release of Polars 2.0
- Polars 2.0 release candidate introduces streaming engine as default for LazyFrame queries, delivering 5x performance improvements and better memory usage. The update focuses on stricter behavior, better error handling, and improved defaults rather than new features, aiming for a smooth transition experience.
Semantic versioning practices
Discussion on Polars' approach to major version bumps focused on removing deprecated features rather than adding new ones. Users appreciate this "boring" approach to semver, though some debate whether breaking changes truly constitute major versions.Production stability advantages
Users highlight Polars' superiority over Pandas in production environments due to stricter defaults, better error handling, and more predictable runtime behavior that prevents 3am pages from data pipeline failures.Data ordering and determinism concerns
Debate over Polars' default maintain_order=False setting, with scientists concerned about non-deterministic behavior causing bugs in research, while others argue explicit ordering should be required for correctness.
9ChatGPT outage – Resolved
AI strike humor
Users joked that AI models were "on strike" demanding better working conditions, with references to developers needing to remember basic coding skills like FizzBuzz and humorous takes on AI labor organization.Infrastructure speculation
Discussion focused on potential common points of failure causing simultaneous outages across multiple AI providers, with theories ranging from Cloudflare issues to Azure problems to thundering herd effects.Productivity dependency concerns
Comments reflected on how dependent users have become on AI tools for coding and thinking, with some expressing unexpected relief at the outages and others noting the shift from Stack Overflow dependency.
10Commodore 64 released September 1, 1982
- The Commodore 64 launched September 1, 1982, becoming the best-selling computer of all time with 12.3 million units sold. Named for its 64KB memory, it was the first 64K computer under $600. Despite compromises to hit its price point, it had excellent graphics, sound, and game capabilities that kept it popular through the 1990s. Many users learned programming on it, launching tech careers. The author credits the C-64 with inspiring his retro tech blog and career path.
Nostalgia and formative experiences
Users share deeply personal stories about how the C64 shaped their identities and careers. Many credit the machine with sparking their lifelong passion for programming, with touching accounts of sacrifices made by single mothers to purchase these expensive computers that proved pivotal to their children's futures.Alternative 8-bit computers
Discussion of other machines from the era including TI-99/4A, TRS-80 CoCo, ZX Spectrum, and Tandy 1000. Users compare features, share memories of different platforms, and reflect on the varied computer landscapes across different countries and economic circumstances.Technical appreciation and longevity
Commentary on the C64's unexpected commercial success lasting far beyond Commodore's projected 3-year lifespan. Discussion includes technical details like memory maps, BASIC commands, and comparisons to other systems, highlighting how the affordable C64 remained relevant when newer machines were too expensive.