Hacker News
1Recent AI model progress feels mostly like bullshit
- An AI startup founder shares their experience noting that despite recent AI model advancements and impressive benchmark scores, they haven't seen meaningful improvements in real-world performance since August 2023. They question whether AI benchmarks accurately reflect practical usefulness and suggest testing models on more complex, real-world tasks.
Benchmark gaming and test data contamination
Discussion of how AI models perform poorly on new math tests (5% on USAMO 2025) despite claims of high performance, suggesting they memorize answers rather than truly solve problems. Concerns about training data contamination.Limited practical improvement in recent models
Mixed experiences with newer models like Claude 3.7 and Gemini 2.5, with some seeing improvements but many noting lateral moves or regressions in practical use cases despite benchmark gains.Models trying to sound smart
Discussion of how LLMs tend to highlight problems and appear knowledgeable rather than give straightforward answers, mirroring human conversation patterns due to their training on language data.
2Rsync replaced with openrsync on macOS Sequoia
- Apple has replaced the aging rsync 2.6.9 with openrsync in macOS Sequoia due to licensing concerns, switching from GPL to BSD-style licensing. While openrsync maintains compatibility with rsync, it supports only a subset of rsync's features, which may affect existing Mac administration workflows that relied on specific rsync functionality.
File copy metadata preservation
Debate about the importance of preserving file metadata during copies. While some users insist on perfect metadata preservation, others, including enterprise backup providers, suggest most users don't care about metadata.GPL licensing concerns
Apple and other companies avoid GPLv3 due to legal ambiguity, potential court cases, and restrictions on device lockdown. Companies prefer permissive licenses like MIT/BSD over GPL's stricter terms.Openrsync limitations
Discussion of openrsync's incomplete feature set compared to regular rsync, particularly lacking support for extended attributes, ACLs, and hard links, causing concern about its viability as a replacement.
3A startup doesn't need to be a unicorn
- A founder shares insights from selling his startup Vizzly to WPP, advocating for a "middle path" in B2B SaaS between VC funding and bootstrapping. He suggests raising under $1M while retaining 90%+ equity can provide optimal returns with less risk than traditional venture-backed or bootstrapped approaches.
German business ecosystem
Discussion of Germany's state support for startups and "Mittelstand" companies, focusing on government grants, bureaucracy challenges, and how this model enables sustainable business growth without VC pressureRegular business vs startup
Debate over what constitutes a "real" startup versus a regular business, with focus on funding approaches, growth expectations, and whether VC-style hyper-growth is necessary for successInvestment returns
Analysis of risk/reward tradeoffs for investors in smaller companies targeting modest growth vs traditional VC unicorn hunting, including discussion of alternative investment options like real estate
4Fewer Foreign Passengers Are Flying to the US
- The author discusses early morning insomnia leading to discovering CBP travel data, revealing a significant decline in foreign travelers entering major US airports compared to last year. Through careful data analysis and validation methods, they confirmed this trend was real and not due to reporting issues, while showing increased US passenger numbers remained stable.
Comment redirection
Discussion redirected to another HN thread about an Axios post, with moderator confirming the move of comments to maintain conversation continuity
5Middle-aged man trading cards go viral in rural Japan town
- In a small Japanese town called Kawara, children are playing a unique trading card game featuring local elderly men (ojisan) instead of typical fantasy characters. Created by community leader Eri Miyahara, the game includes 47 cards showcasing local men with special abilities based on their real-life contributions, fostering stronger connections between generations and increasing community participation.
Community building through trading cards
Trading card game featuring local elderly men strengthens intergenerational bonds, doubles community event participation, and creates real-world role models for kids while preserving elders' stories and contributions.Japanese vending machines
Discussion of Japan's innovative vending machine culture, from hot/cold drinks to unique items. Debate about how Japanese innovations like cold coffee were initially dismissed but later adopted globally.Gender representation
Debate about the absence of women in the trading card project, with responses ranging from product development strategy to addressing specific social issues like male loneliness in Japan.
6Let's Ban Billboards
- A call to ban billboards in cities, highlighting the contrast between strict design review processes for buildings and the unregulated placement of large, distracting advertisements. The author argues that removing billboards would create a more peaceful urban environment benefiting everyone except a few landowners.
Places with billboard bans
Several states (Vermont, Alaska, Hawaii, Maine) and cities have banned billboards, with residents noting the dramatic positive impact on visual environment and quality of life when crossing between banned/allowed areas.Impact on environment and attention
Billboards create visual pollution, distract drivers, and make areas feel "mentally oppressive." Their absence makes places feel quieter and more natural, with highways better blending into surroundings.Economic and regulatory aspects
Discussion of billboard ROI, advertising costs, local regulations, and revenue implications. Some areas restrict ads to local businesses only, while others debate complete bans vs incremental restrictions.
7Capitol Trades: Tracking Stock Market Transactions of Politicians
- Data accessibility on the website is limited to a 3-year historical period.
Strict trading restrictions needed
Discussion emphasizes need for tighter controls on Congressional trading, with many arguing current 30-day disclosure period is inadequate. Debate over whether blind trusts or index funds should be required.Trading performance analysis
Evidence suggests Congressional trading doesn't consistently beat market averages, contradicting popular belief. Focus on Pelosi's trades draws criticism as other politicians show similar or better returns.Investment restrictions debate
Arguments over whether public officials should be banned from stock ownership entirely vs allowing index funds. Discussion of conflict of interest concerns and appropriate investment vehicles.
8How the Atlantic's Jeffrey Goldberg Got Added to the White House Signal Chat
- Trump's national security adviser Mike Waltz accidentally added journalist Jeffrey Goldberg to a Signal group chat about Yemen strikes due to a phone contact mix-up. Trump considered firing Waltz but decided against it to avoid giving satisfaction to the media, particularly The Atlantic magazine.
Contact auto-suggestions
Discussion of how iPhone's contact suggestion feature led to the leak, with many debating whether this explanation is credible and highlighting risks of auto-suggestions in sensitive communicationsUse of Signal vs secure systems
Debate about why officials used Signal instead of secure government networks, with some citing convenience and cross-agency limitations while others argue proper systems exist but were deliberately avoidedResponse to the leak
Discussion of how the administration handled the leak aftermath, with many criticizing the "cleared" narrative and arguing it demonstrates reckless behavior rather than exoneration
9Show HN: Browser MCP – Automate your browser using Cursor, Claude, VS Code
- Browser MCP is a tool that enables AI applications to automate browser tasks by connecting them directly to your browser. It allows for automated form filling, testing, and workflow automation while maintaining security and privacy since operations happen locally. The tool offers various browser control features and works with popular AI applications.
Browser automation challenges
Discussion of limitations with browser automation, including CAPTCHA triggers, detection by CloudFlare, and how timing/mouse movements can reveal automated behavior. Users share experiences with blocks and suggestions for mimicking human behavior.Understanding MCP
Debate about what MCPs (Model Context Protocol) are, with explanations of it being an API layer allowing LLMs to interact with browsers and other tools. Discussion includes comparison to RPA and potential security concerns.Specific use cases
Users sharing desired applications for browser automation, including shopping with specific criteria, reimbursement requests automation, and controlling hardware synthesizers. Discussion of limitations in complex UIs like Google Sheets.
10Glamorous Toolkit
- Glamorous Toolkit is a moldable development environment that enables developers to create contextual micro-tools for explaining and exploring systems. It supports multiple languages and technologies, allowing users to build custom tools for specific problems while making systems more understandable through interactive visual experiences.
Project clarity and approachability
Multiple users found GT difficult to understand from website/docs. Discussion focused on need for better explanation of use cases, less academic language, and clearer onboarding path vs current reliance on videos and complex terminology.Comparison to existing tools
Debate around how GT compares to Jupyter notebooks and other dev tools. Users discussed unique features like live objects and moldable development, while questioning whether benefits justify switching from familiar environments.Learning curve challenges
Users expressed frustration with steep learning curve and Smalltalk environment. Discussion covered need for better documentation, self-explanatory basic functionality, and easier path to productivity for newcomers.