2026 06 09 HackerNews

2026-06-09 Hacker News Top Stories #

The author, who experienced addiction, incarceration, and a felony charge from the age of 14, eventually rebuilt a stable life and career through the open-source community and proactive efforts. "Dopamine Fracking" refers to the process of extracting pure pleasure from experiences through capital and algorithms, leading to cultural homogenization and the disappearance of authentic experiences. performative-ui is a React component library for building frontend prototypes for AI products, wrapping common marketing design tropes in a humorous and self-deprecating manner. Social media is shifting from friend-based interactions to algorithm-driven popular content, making users passive consumers while advertising revenue continues to grow. Linear's speed is key due to its local-first architecture, caching data in IndexedDB on the browser side and asynchronously syncing it, achieving instant UI responsiveness. Xiaomi's MiMo-V2.5-Pro-UltraSpeed achieves over 1000 tokens per second decoding speed on a 10 trillion parameter model, using techniques like quantization to achieve ultra-fast inference on commercial GPUs. A farmer family in the US donated land to the city for use as a park, but the city illegally sold it to a data center developer; the donors' lawsuit was dismissed by the court. Over 450 images in Thermo Fisher Scientific's antibody catalog were found to have manipulation traces such as copying and flipping, suggesting systematic data fabrication. DeepSeek V4 Pro outperforms GPT-5.5 Pro with a score of 38.0 to 33.0 in accurately executing four text tasks, mainly due to strict adherence to instructions and precise output. xAI rents a large amount of GPU capacity to Anthropic and Google, charging high monthly fees, making its business model more similar to a data center real estate trust than a pure frontier laboratory.

1. Building from Zero After Addiction, Prison, and a Felony #

https://gavinray97.github.io/blog/building-from-zero-after-addiction-prison-felony

At 14, the author was incarcerated for two years due to drug addiction and drug trafficking, and obtained a GED in a maximum security juvenile prison. After release, he briefly worked and attended a community college, but relapsed into drug use and trafficking, becoming a felony offender at 18. During his time in county prison, he saw a news story about a tech company offering internships to disadvantaged youth, applied, and was hired as a full-stack web development intern. There, he met his future wife, but again fell into drug abuse and lost his job. He then moved to Florida to live with his father, but the situation worsened. Eventually, through open-source contributions and community involvement, he rebuilt his career, hoping his experience would encourage others in similar situations.

HN热度 876 points | 评论 395 comments | 作者：gavinray | 1 day ago #

https://news.ycombinator.com/item?id=48437406

• People with non-traditional paths have entered the tech industry through free internships and now have stable lives and families, proving that it's possible to start over from the bottom.

• Someone summarized this type as "34-year-old Patrice," with a criminal record but no one knows about it after ten years, allowing a fresh start in life.

• A colleague was incarcerated for a felony related to growing marijuana and now has a normal work and family life after several decades.

• Society's moral standards regarding drugs are capricious; what was once a crime is now entrepreneurship, and people should think independently rather than blindly follow society.

• People with similar experiences moved to Sri Lanka and found that employers do not discriminate based on drug records.

• However, some argue that employers who check records will choose candidates without any records under equal conditions.

• Government security reviews focus more on financial debt, gambling, and suspicious relationships, with short-term drug use history not being a priority.

• Job application forms have changed "bad behavior" from drug use to traffic violations, showing cultural differences.

• A drug history from the distant past with a stable life is not a big issue, but recent drug use remains an obstacle.

• In the Bay Area tech scene, drug use (like at Burning Man) might actually bring career opportunities.

• Undisclosed drug records can still be used against you.

• Asking how to prevent young people from taking this path, the current fentanyl crisis makes this path even more dangerous.

• Replies emphasize that education provides better options, but the government should not prevent personal choices unless they harm others.

2. Dopamine Fracking #

https://igerman.cc/blog/dopamine-fracking/

"Dopamine Fracking" — the author proposes a concept: investing a large amount of resources (money, data, algorithm optimization, etc.) into activities that were originally casual and complex, forcibly extracting the purest dopamine pleasure, regardless of long-term damage. Like real hydraulic fracturing, this practice has a destructive impact on culture, creativity, and interpersonal relationships.

The author uses the example of strawberries: industrial extraction of strawberry flavor compounds replaces the complex taste, texture, and emotional memory of real strawberries, eventually leading to the disappearance of the original strawberry. This phenomenon occurs in various fields such as music, film, and online communities — everything is aimed at pursuing dopamine peaks, leading to cultural homogenization and impoverished experiences.

Finally, the author shares personal coping methods: identifying and closing channels and content that deliberately stimulate dopamine, and regaining real and rich experiences.

HN热度 748 points | 评论 382 comments | 作者：igmn | 21 hours ago #

https://news.ycombinator.com/item?id=48440792

• Platforms like YouTube create low-quality, manipulative content (such as child-inappropriate collages, emotional manipulation) to pursue profits, sacrificing user well-being.

• Parents should not let young children watch YouTube alone, as the content has seriously deteriorated into pursuing clicks rather than quality.

• Parents are forced to completely disable YouTube or strictly limit it, which is a necessary measure to protect children's brains from excessive dopamine stimulation and addiction.

• Even teenagers should not be exposed to this garbage content, and YouTube needs high-level screening to be valuable.

• Deceptive advertisements (such as fake investment) have existed for a long time, and platform regulation is weak.

• Parental negligence and platform addiction mechanisms jointly cause the problem, and it cannot be solely blamed on either party.

• Products designed as "torture hubs" for children should not be accepted by society.

• Playing YouTube videos in kindergartens should also be banned.

• Modern media is intentionally designed to be addictive, different from the 1990s, and parents need to be more proactive in participating and setting usage rules.

• Tech companies create addiction and shift responsibility to parents, which is malicious.

• Many parents are also addicted to screens, leading to young children being exposed to screens too early, which is harmful.

3. Show HN: Performative-UI – A React Component Library of Design Tropes #

https://vorpus.github.io/performativeUI/

performative-ui is a React component library for AI-native products in the forefront of laboratories, containing 27 MIT-licensed components. The page humorously and self-deprecatingly displays the classification of each component:

• Atoms: Sparkle, GradientText, StatusDot, etc., basic elements.

• Primitives: Button, EyebrowPill, Prompt, etc., interactive primitives.

• Banners: StickyBanner, used for disguising funding news as practical information.

• Heroes: Rotator, WordRoll, PromptHero, AsciiHero, used for showcasing product breadth or alternative value propositions.

• Backgrounds: Aurora, NodeGraphBackground, FloatingSparkles, etc., visual effects backgrounds.

• Surfaces: GlassCard, MockIDE, simulating real code or glass-like interface.

• Conversation: ChatBubble, TokenStream, ChatFAB, realizing real-time conversation streams and floating chat buttons.

• Social Proof: LogoMarquee, LogoRow, StatCounter, CommunityBadge, used for displaying user numbers, partners, and community heat.

• Pricing & Conversion: PricingCard, BeforeAfter, WaitlistForm, Popover, used for pricing display, conversion funnel, and waitlist.

Component descriptions are filled with self-deprecating marketing language (such as "buttons move to make you click," "numbers going up are better than not going up"), and the overall style is light-hearted and mocking, suitable for quickly building frontend prototypes for AI products.

HN热度 713 points | 评论 145 comments | 作者：lizhang | 9 hours ago #

https://news.ycombinator.com/item?id=48445554

• Websites without performative UI are seen as unprofessional.

• Annoying popups and subscription buttons, although annoying, are effective.

• Clean and simple UI gives a reliable first impression.

• Some websites (like Craigslist, Temu) have basic UI but are still successful, showing that functionality is more important.

• Homogenization of internet UI is inevitable, and general-purpose frameworks reduce the user's learning cost.

• Homogenized tools can reduce cognitive load, allowing users to focus on differentiated content.

• Too intuitive UI may cause users to skip learning, while more complex UI forces higher focus.

• Migrating to Material UI or building a custom component library involves trade-offs, and the decision should be based on needs.

• Performative UI is a social signal, indicating that the project belongs to a successful trend group.

• Design languages change over time, and adopting the latest trends implies that the website is new or up-to-date.

4. Anti-social: Now, fads, not friends, dominate social media feeds #

https://www.bbc.com/worklife/article/20260520-how-social-media-ceased-to-be-social

Social media is shifting from communication between friends to an entertainment platform. Many users no longer post frequently but passively watch professional content. Data shows that the proportion of active posting users has decreased in France, the UK, and the US, especially among Gen Z, with only 18% active and 74% passive.

Users are turning to private groups and messaging apps like WhatsApp to avoid ads and influencer content. Algorithms on platforms like Instagram and Facebook are increasingly favoring content from strangers over friends' updates to increase user engagement and ad revenue.

Small business owners are forced to also act as content creators to gain visibility. Despite changes in user behavior, advertising revenue on social platforms is still growing, expected to reach $317 billion globally by 2026.

HN热度 514 points | 评论 378 comments | 作者：1vuio0pswjnm7 | 12 hours ago #

https://news.ycombinator.com/item?id=48444228

• Social media is now like cable TV in the past, but even worse; it's purely designed to manipulate you, making you feel unsafe and using your emotions for others' purposes.

• Hacker News is not typically considered social media, although it has a front-end ranking algorithm, it is relatively transparent and does not have aggressive algorithmic pushes.

• TV news constantly feeds you the message that "you should be afraid" and "you should be angry," and social media does the same; some people have left Facebook and Reddit due to the negative impact on mental health, and YouTube algorithms also push content that appears rational but actually sells anxiety.

• An Italian immigrant watches Fox News criticizing "chain migration," while his family succeeded through chain migration, and this contradiction is frustrating.

• Older immigrants often have an attitude of "taking the ladder away" towards later immigrants, claiming "they should follow the rules like I did," but the rules back then were much simpler.

• Historically, changes in immigrant attitudes are related to the proportion of immigrants: currently, the proportion of foreign-born people in the US is close to 15.8%, and anti-immigrant sentiment is repeating the history of the late 19th century.

• Every immigration wave has similar arguments against it (taking jobs, not integrating), but in the long run, these are not problems; many people who oppose "illegal immigrants" are actually also dissatisfied with legal immigrants, and current policies (deporting colored refugees, favoring South African whites) confirm this.

• If someone believes that their quality of life has declined due to immigration, they need to provide more specific reasons.

• With global birth rates declining and a need for a labor force, the US wants to shrink its population at an inopportune time.

• Many people are ignorant of the rules and the actual situation, but the idea of "following the rules" is deeply rooted; people can get angry about minor rules (like toilet water volume, fare evasion) but also secretly break them.

5. How’s Linear so fast? A technical breakdown #

https://performance.dev/how-is-linear-so-fast-a-technical-breakdown

Linear is a project management tool, and its speed is extremely fast, due to its technical architecture design: the core is using IndexedDB as a local database in the browser, and user operations directly update local data, then asynchronously synchronize to the server and broadcast to other clients via WebSocket, eliminating network waiting time. This "local-first" model combined with MobX's reactive updates makes the UI respond instantly without a loading state.

In terms of first load optimization, Linear continuously iterates its build tools (Parcel → Rollup → Vite → Rolldown), significantly reducing the size of JavaScript/CSS (reduced by 30% after compression, cold cache loading is 10-30% faster), and abandoning support for older browsers to eliminate polyfills and ES5 transpilation. The overall technology stack is simple: frontend React + MobX, backend Node.js + PostgreSQL, all code uses TypeScript, and it sticks to client-side rendering (CSR) instead of server-side rendering to reduce complexity.

The article emphasizes that for most web applications, using optimistic updates (such as TanStack Query, SWR) can significantly improve the experience, and the core principle is that UI responsiveness should not be limited by network latency.

HN热度 476 points | 评论 227 comments | 作者：howToTestFE | 1 day ago #

https://news.ycombinator.com/item?id=48437609

• After using Linear in practice, enthusiasm was lost, search is slow, UI is clumsy, Pulse is noisy, and it's not as good as the early Trello.

• Linear's marketing has created an image of being "fast," but the actual experience is average, and many people recommend it without having used it themselves.

• It's not enough to just be Jira plus dark mode and a usable API to be successful.

• After using Linear for two years, I think the hype is unreasonable, and going back to Jira won't affect efficiency.

• Early Linear was indeed fast, but now it's bloated with features, shifting from small companies to mid to large enterprises.

• Whether or not performance matters depends on the role: only those who use Jira all day care about speed.

• Self-hosted Jira might be better than Atlassian's shared hosting.

• Self-hosted Jira has countless small issues, which are worse than the hosted version.

• Linear is objectively better than Jira, and there's no comparison.

• Linear used to be simple and fast, but it has gradually become bloated with features and UI has become messy.

• Making your own task tracker with native OIDC, status comments, API batches, etc., is better than commercial products.

• UI is messy, and core functions are buried under too many features.

• When opening tickets during meetings, it takes a long time to load the cache, which is embarrassing.

• GitHub issues loaded faster before the React rewrite.

• Middle-clicking to open a new tab often gets stuck in the loading state, and even a forced refresh doesn't help.

• The common weakness of SPAs is that links opened in new tabs are prone to being stuck.

• Linear may be starting to "shit," becoming bloated with features and aggressively chasing the AI trend.

• Trello is still useful, and I hope it doesn't get shit.

• It's possible to clone a Trello in 30 minutes with vibe code, but quality and usability are hard to guarantee.

• The premise of local-first synchronization is problematic: traditional CRUD can also achieve 30ms response time if it's close to the user's backend.

• 300ms is not fast, 30ms is the target, and every bit of JS has a cost.

• "Sub-second" response used to be the goal, but now the standard is higher.

• Being close to the user requires users to be concentrated or only cater to US users, otherwise the complexity is very high.

• A decade ago, a project used PHP SSR to achieve 30ms response time, covering 80% of the global population with 3 AWS regions in less than 100ms, it's interesting to solve old problems with new methods now.

6. MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second #

https://mimo.xiaomi.com/blog/mimo-tilert-1000tps

Xiaomi and TileRT have jointly launched the MiMo-V2.5-Pro-UltraSpeed, achieving a decoding speed of over 1000 tokens/s (up to approximately 1200 tokens/s) for the first time on a 10 trillion parameter model. This API is temporarily open (from June 9 to June 23), and access is by application. The price is three times that of the standard version, but the speed is about ten times faster. This breakthrough in speed brings about a paradigm shift: multiple inference paths can be run in parallel within the same time frame, improving efficiency in scenarios such as code generation and real-time decision-making (e.g., high-frequency trading and medical image analysis). Technically, it adopts FP4 quantization (only for MoE expert layers) and DFlash speculative decoding, among other model-system co-design approaches, achieving ultra-fast inference on standard 8-card commercial GPUs.

HN热度 466 points | 评论 316 comments | 作者：gainsurier | 8 hours ago #

https://news.ycombinator.com/item?id=48446639

• Fast AI is extremely fast, which may change work processes, exciting yet unsettling, leaving people unsure of what humans can still do.

• AI's estimated time (e.g., "part 2-3 weeks") is often inaccurate, with actual tasks completed in half an hour, possibly due to advertising or hallucination.

• AI's time estimation comes from training data (e.g., Jira, GitHub tasks) or is intentionally trained by product teams, not from actual capabilities.

• Modern AI systems (e.g., Codex, Claude Code) have surpassed pure LLMs, incorporating agents, sub-agents, memory, and tools, with reasoning modes close to human.

• Some argue that modern systems are still context engineering and UI packaging, fundamentally the same as in 2024, not revolutionary changes.

• AI can automatically perform experiments, observe results, and iterate in programming, similar to the debugging process of human engineers.

• AI's estimation should be improved based on real time data from chat logs, not relying on unreliable numbers generated by models.

• Even if AI quickly completes the code, subsequent testing, review, and patching still take a long time, and the total time may be the same as human estimation.

7. A Farmer Donated Land to Turn into a Park. The City Is Building a Data Center #

https://www.404media.co/a-farmer-donated-land-to-turn-into-a-park-the-city-is-building-a-massive-data-center-instead/

In 1999, a farmer's family donated 87 acres of land to the city of Tyler, Texas, for a nominal fee of $10, with the condition that the land must be used as a public park. In 2025, the city sold the land for $10 million to data center developer Blueprint, planning to build a 135,000 square foot data center. The Griffin family, who have lived in the area for generations, used to play and camp on the land, and now the data center will be built just 500 feet from her home.

HN热度 410 points | 评论 226 comments | 作者：greedo | 8 hours ago #

https://news.ycombinator.com/item?id=48446439

• The farmer donated the land with a restriction for park use, but the city sold it to a data center, and the family's lawsuit was dismissed.

• Neighbors lack standing to sue, as only the original landowner or the city can enforce the covenant restrictions.

• As residents and taxpayers, neighbors face property devaluation and should have the right to sue.

• Indirect damages typically do not grant standing to sue, requiring direct harm for a lawsuit.

• The city has the right to release covenant restrictions if it serves the public interest, but this does not necessarily mean it is morally right.

• To ensure park use, protective easements or return clauses should be established.

• Once land ownership is transferred, the original owner cannot permanently restrict its use, requiring legal tools to ensure it.

8. How much of Thermo Fisher’s antibody data has been manipulated? #

https://reeserichardson.blog/2026/05/28/how-much-of-thermo-fishers-antibody-data-has-been-manipulated/

This article reveals that Thermo Fisher Scientific has manipulated a large number of images in its online antibody catalog. Researchers Sholto David and Reese Richardson found that over 450 Western blot images used to demonstrate antibody specificity showed signs of manipulation, including copying, flipping, rotating, and tampering. One background noise pattern was even repeated in the validation data of more than 50 different antibodies. These manipulations make scientists unable to trust the actual performance of the products, while the antibodies are expensive (a small vial costs $400–$500). Thermo Fisher responded that image adjustments were only for "optimizing display clarity," but the authors argue that this cannot explain the obvious forgeries. The article calls on scientists to verify antibodies themselves and invites the public to report suspicious images.

HN热度 388 points | 评论 85 comments | 作者：mhrmsn | 17 hours ago #

https://news.ycombinator.com/item?id=48442075

• Comments argue that this is a clear case of systemic fraud, similar to past data falsification cases, but it's important to distinguish between image fraud and research data fraud.

• It points out that image compression algorithms (e.g., JBIG2) may lead to incorrect replacement of digital data, causing serious consequences, but this issue is unrelated to the situation in the article.

• Modern phone AI enhancement may fabricate details (e.g., Samsung's moon photos), raising concerns about the concept of data quality being diluted.

• Praise is given to whistleblower Sholto David, who received millions in rewards for reporting fraud.

• Whistleblowers can share in government fines, and short sellers often profit from this route as well.

• Some labs have already discovered similar antibody data fraud in the past and thus actively avoid certain suppliers.

• Such systemic fraud wastes a lot of money and time, leading to multiple paper retractions and having a significant impact.

• Scientific image processing should be an automated process, and deliberate forgery requires higher expertise and is difficult to conceal.

• The explanation that "sales managers use images without original images" is almost impossible to be true.

9. DeepSeek V4 Pro beats GPT-5.5 Pro on precision #

https://runtimewire.com/article/deepseek-v4-pro-beats-gpt-5-5-pro-on-precision

In the latest round of AI model competition, DeepSeek V4 Pro defeated GPT-5.5 Pro with a score of 38.0 to 33.0, mainly due to its precise execution of tasks. Four text tasks were tested: log sanitization (DeepSeek used a single regex to handle overlapping patterns more reliably), supplier delay notification (strictly followed instructions without adding extra content), meeting minutes JSON (precisely matched schema), and messy order to JSON (both tied). Conclusion: DeepSeek V4 Pro is more rigorous and reliable, especially showing a clear advantage in tasks where minor deviations can lead to failure.

HN热度 387 points | 评论 214 comments | 作者：yogthos | 22 hours ago #

https://news.ycombinator.com/item?id=48440448

• The article's experimental design is sloppy and lacks rigor, similar to AI-generated clickbait.

• Criticism of the lede's sensational language: the lede's intention is to attract readers, and it should not be overly criticized.

• Debate on the origin of the term "lede."

• Discussion on whether the article's goal is noble: comparing models is a legitimate goal, but the methodology is problematic.

• There is a motive to not want cheap Chinese models to succeed.

• Chinese models are cheap because they are derived from others' creations.

• Under the spirit of open source, all models are created collectively by everyone, and there should be no double standards.

• An analogy to the automotive industry explains that models have different applicable scenarios, and there is no absolutely best model.

• Blindly rejecting based on intuition is low quality.

• The tasks selected favor DeepSeek, not fully showcasing GPT-5.5 Pro's strengths.

• The test results are consistent with other instruction-following benchmarks (ifbench), and have some reference value.

• The Pelican test also lacks rigor but is not criticized in the same way.

• The Pelican test is essentially a fun curiosity and should not be held to scientific rigor standards.

• The Pelican test has been criticized multiple times, but the author chooses to ignore it.

• The announcement of the Pelican test's death is based on strange standards (Qwen surpassing Opus).

10. xAI looks more like a datacentre REIT than a frontier lab. #

https://martinalderson.com/posts/xais-new-rental-business/

xAI has reached large-scale datacentre capacity leasing agreements with Anthropic and Google, charging up to $1.25 billion per month (300MW) and $920 million per month (110k GPU), respectively, generating substantial profits. Anthropic previously had to limit usage during peak times due to a shortage of computing power, and the pressure was alleviated by xAI's Colossus 1 datacentre. Although some question whether these transactions might be a means for Musk to pressure OpenAI or inflate the IPO valuation of SpaceX, the global shortage of GPUs is real, and xAI (affiliated with SpaceX) has a competitive advantage in the rapid construction of datacentres (such as building Colossus 1 in 122 days). However, this has led to the computing power originally used for Grok training and inference being rented out to competitors, potentially impacting Grok's position as a frontier model. Overall, xAI is increasingly resembling a datacentre REIT with a frontier lab, rather than a purely frontier lab.

HN热度 357 points | 评论 270 comments | 作者：martinald | 8 hours ago #

https://news.ycombinator.com/item?id=48446428

• xAI is more like a datacentre REIT than a frontier lab, and the cyclical transactions with SpaceX and Google raise doubts

• Some believe that the scarcity of computing resources is a fact, and xAI is managing it well, not engaging in cyclical transactions

• Computing assets are depreciating rapidly, similar to a car rental company, where idle GPUs are more costly than cars

• Opposing view: GPUs are appreciating due to strong demand, with old cards increasing in price from $600 to over $1,000

• Frontier labs are shifting from pricing based on computing power to pricing based on intelligent value, increasing computing costs and squeezing open-source model competition

• Evidence of "dark GPUs" and paused datacentre construction may indicate a bubble

• Others point out that the main reason for the pause in construction is urban resistance and power shortages, not insufficient demand

• Analogy to dark fiber in the 1990s: fiber optic technology iteration allows old fiber to be expanded, but GPU upgrades are different

• xAI gains electricity by rapidly deploying GPUs and gas turbines, although the environmental impact is questionable, it has won a first-mover advantage

• Hardware prices have risen several times due to AI demand, and it is wise to quickly monetize them before depreciation

• xAI allows Google to act quickly through an intermediary company while maintaining distance

Hacker News精彩评论及翻译 #

Anti-social: It's fads, not friends, which now dom… #

https://news.ycombinator.com/item?id=48444940

Social media is now exactly what cable television used to be, but worse; it exists solely to coerce you. It makes you feel insecure, and it leverages your emotions for someone else's aims.

Due to the changes in technology, social media is far more effective at this than cable TV ever could be, but the concept is the same. It's some remote person attempting to manipulate you by also packaging something enticing along with that manipulation. It's long past time to leave it permanently.

And no, HN is not social media in any normal sense of the word. The pedantry involved in that comparison is extremely tiresome.

everdrive

社交媒体现在完全就是过去的有线电视，但更糟；它的存在只是为了胁迫你。它让你感到不安，利用你的情绪为他人达成目的。由于技术变革，社交媒体在这方面远比过去的有线电视更有效，但概念是相同的。某个遥远的人试图操纵你，同时将诱人的东西与这种操纵打包在一起。早就该永久地离开它了。不，HN不是任何正常意义上的社交媒体。那种比较中的迂腐极其令人厌烦。

A Farmer Donated Land to Turn into a Park. The Cit… #

https://news.ycombinator.com/item?id=48447117

Wow they had the condition that the land be used as a park baked into the deed when they sold it to the city for $10, the city sold it, and when the family went to court their suit was dismissed. Now their home is worthless because nobody wants to live next to a data centre.

When are we going to hold local government officials accountable for bullshit like this? Send them to prison.

helterskelter

哇，他们在以10美元卖给城市时，把这块地必须用作公园的条件写进了契约，结果城市又把它卖了，而这家人在法庭起诉时被驳回了。现在他们的房子一文不值，因为没人想住在数据中心旁边。我们什么时候才能追究地方政府官员这种胡作非为的责任？把他们送进监狱吧。

Apple WWDC 2026 #

https://news.ycombinator.com/item?id=48448393

Apple very rarely admits mistakes. The fact they're rolling back some of the extremeness in Liquid Glass and actively mentioned in the keynote that they very seriously took the user feedback shows just how bad it was, at least initially.

WoodenChair

苹果很少承认错误。他们正在撤回Liquid Glass中的一些极端设计，并在主题演讲中主动提到他们非常重视用户的反馈，这一事实表明至少最初这个问题有多严重。

https://news.ycombinator.com/item?id=48448476

They try so hard to do a polished presentation that everything is kinda fake and unauthentic. I don't understand how this attitude survived so many years.

antirez

他们如此努力地做一个光鲜的展示，以至于一切都显得有点虚假和不真实。我不明白这种态度怎么能持续这么多年。

Building from zero after addiction, prison, and a … #

https://news.ycombinator.com/item?id=48439600

Had similarly unorthodox path to tech, albeit without the drug addiction or prison.

90s early internet/BBS punk rocker/computer nerd. Hated school angry.

Dropped out to work as a bike messenger for 5 years before packing a bag and moving west randomly. Couldn’t sit still. Rode freight trains around the country for a few months.

Washed dishes and landscaped to cover my cheap rent till that fell thru. Discovered shop lifting. Covered food and beer stealing from local progressive grocery store chain. Stole goods to sell on CL to cover my rent. That scam went tits up and narrowly escaped serious charges after the head of loss prevention from a regional retailer caught up to me

Was sleeping in the park–this was pre super meth/fentanyl crisis so street living was a bit more stable and low key. Didn’t want to wash dishes or dig holes any more so looked around on CL. Found a small company trying to bootstrap a regional office for an established linux-related open source company. Worked for free / interned using a stolen laptop for a year or so while sleeping outside or couch surfing local punk houses.

Eventually got hired on for s but stayed for a couple years and made many FOSS connections. Eventually left to join a well known FOSS-centered company that was fully remote.

Told myself when I was young that I would never work in an office. ~15 years later and I never have ,but now work in bit tech, get paid too much, own a home and have a great family with kids who play at the same parks I used to crash at. We shop (and pay) at the same stores I used to crib from.

I’m respected and tenured at my gig but Imposter syndrome still holds me back. Nobody I work with knows where I came from and thankfully have nothing incriminating that would block a background check

mapassthebeans

Similarly entered tech in a non-traditional way, but without a drug use or incarceration history.

90s early internet/BBS punk rocker/computer nerd. Hated school angry.

Dropped out of school to work as a bike messenger for 5 years, then randomly packed a bag and moved west. Couldn't sit still. Rode freight trains around the country for a few months.

Washed dishes and did landscaping to cover my cheap rent until that fell through. Discovered shoplifting. Stole food and beer from a local progressive grocery store chain to cover my expenses. Sold stolen goods on Craigslist to pay my rent. That scam went south, and I narrowly escaped serious charges after being caught by the head of loss prevention from a regional retailer.

I was sleeping in the park—at that time, before the super meth/fentanyl crisis, street living was a bit more stable and low-key. I didn't want to wash dishes or dig holes anymore, so I looked around on Craigslist. I found a small company trying to bootstrap a regional office for an established Linux-related open source company. I worked for free or interned using a stolen laptop for a year or so, sleeping outside or couchsurfing in local punk houses.

Eventually got hired on, but only stayed for a couple of years and made many FOSS connections. Eventually left to join a well-known FOSS-centered company that was fully remote.

Told myself when I was young that I would never work in an office. ~15 years later and I never have, but now work in big tech, get paid too much, own a home, and have a great family with kids who play at the same parks I used to crash at. We shop (and pay) at the same stores I used to crib from.

I’m respected and tenured at my gig, but Imposter syndrome still holds me back. Nobody I work with knows where I came from, and thankfully, I have nothing incriminating that would block a background check.

LLMs are eroding my software engineering career an…

https://news.ycombinator.com/item?id=48435521

I don’t know what to do.

Ride the wave. You rode it when websites/webapps were the wave. I came into the software industry before the internet, kept changing my horse. You are never too old to learn new tricks. The new wave creates new kinds of work and workers. Be one of them. Ride the beast, master the tools. It’s the same game again.

zkmon

I don’t know what to do.

Ride the wave. You rode it when websites/webapps were the wave. I came into the software industry before the internet, kept changing my horse. You are never too old to learn new tricks. The new wave creates new kinds of work and workers. Be one of them. Ride the beast, master the tools. It’s the same game again.

Anthropic, please ship an official Claude Desktop …

https://news.ycombinator.com/item?id=48434916

Hey!

I manage the unofficial build at https://github.com/aaddrick/claude-desktop-debian

Debian is in the name, but the scope has grown to all backends, compositors, etc.

The main reason most companies don’t publish Linux electron apps is fragmentation. If you’re doing anything more than rendering a webpage as an app, it starts to get complicated. I’ve got a bank of VMs set up for testing, and I still need it up.

aaddrick

Hey!

I maintain an unofficial build version, available at https://github.com/aaddrick/claude-desktop-debian

Although the name includes Debian, it has now expanded to support all backends, synthesizers, and so on.

The main reason most companies don't release Linux versions of Electron applications is due to fragmentation. If you're doing anything more than rendering a webpage as an application, things get complicated. I've prepared a bunch of virtual machines for testing, but I still need to keep them updated and maintained.

https://news.ycombinator.com/item?id=48435661

The main reason most companies don't publish Linux Electron apps is fragmentation. If you're doing anything more than rendering a webpage as an app, it starts to get complicated.

Can confirm. At a past company, we worked hard to release a Linux desktop client for our customers who wanted it, even though the number was small.

It turns into compatibility hell very fast. You think you can target a couple of recent Ubuntu releases and everything will be good, but then you're getting peppered with complaints from people using distributions you've never heard of because some part of the app isn't working right. So your engineers spend a half day installing that in a VM and debugging it, but the problem is in upstream somewhere. The number of tickets with Linux issues keeps growing and each one is taking more time to debug, all for a number of customers that is so small you can't justify doing it.

But those customers are angry. And vocal. They're posting all over Twitter, Hacker News, and Reddit about how your company's software is garbage, without mentioning that they're running an unknown distribution on a 13 year old ThinkPad.

This even impacts open source projects. Several popular OSS Electron apps don't work on many popular distros unless you set some command line workarounds, and even then it's flakey. The open source projects get a pass because it's open source, but if your company releases something you might be picking up a lot of angry, vocal customers that you didn't want.

Aurornis

The main reason companies don't publish Linux Electron apps is fragmentation. Once you're doing anything more than rendering a webpage as an app, things get complicated.

Can confirm. At my previous company, we went to great lengths to release a Linux desktop client for our customers who wanted it, even though the number was small.

It turns into compatibility hell very fast. You think you can target a couple of recent Ubuntu releases and everything will be good, but then you're getting peppered with complaints from people using distributions you've never heard of because some part of the app isn't working right. So your engineers spend a half day installing that in a VM and debugging it, but the problem is in upstream somewhere. The number of tickets with Linux issues keeps growing and each one is taking more time to debug, all for a number of customers that is so small you can't justify doing it.

But those customers are angry. And vocal. They're posting all over Twitter, Hacker News, and Reddit about how your company's software is garbage, without mentioning that they're running an unknown distribution on a 13 year old ThinkPad.

This even impacts open source projects. Several popular OSS Electron apps don't work on many popular distros unless you set some command line workarounds, and even then it's flakey. The open source projects get a pass because it's open source, but if your company releases something you might be picking up a lot of angry, vocal customers that you didn't want.

Are you expected to run five Python type-checkers … #

https://news.ycombinator.com/item?id=48446048

If you are going to be super-strict with type-checking, wouldn't it be best to switch to a statically typed language and get the performance gains as well?

dsign

If you're going to be super-strict with type-checking, wouldn't it be best to switch to a statically typed language and get the performance gains as well?

The EU Open Source Strategy #

https://news.ycombinator.com/item?id=48443682

I wish there was somewhere I could earnestly and intelligently have discussions about EU related tech and tech policy, but HN isn't it. As you can see already in this thread, there's 14 comments besides mine and they are 100% negative, and about 95% low effort/reactionary.

Of course there's a lot to criticize and also to appreciate about the EU. But this is supposed to be a forum for intelligent, thoughtful discussion and yet as soon as the EU gets mentioned it basically turns into reddit.

esperent

I wish there was somewhere I could earnestly and intelligently have discussions about EU related tech and tech policy, but HN isn't it. As you can see already in this thread, there's 14 comments besides mine and they are 100% negative, and about 95% low effort/reactionary.

Of course there's a lot to criticize and also to appreciate about the EU. But this is supposed to be a forum for intelligent, thoughtful discussion and yet as soon as the EU gets mentioned it basically turns into reddit.

DeepSeek V4 Pro beats GPT-5.5 Pro on precision #

https://news.ycombinator.com/item?id=48441125

It’s four poorly constructed arbitrary experiments which say very little about the competency of either model.

The article reads like thin, auto-generated AI clickbait for nerd sniping or shilling a model.

Consider the lead:

DeepSeek V4 Pro wins this head-to-head by being more exact where it matters: following instructions, matching schemas, and solving edge cases cleanly. GPT-5.5 Pro is still strong, but it gave away points with avoidable deviations.

“where it matters”, “cleanly”, “is still strong”, and vague references instead of telling 3 out of 4 tests Deepseek yielded more concise results.

1 star.

Stitch4223

These are four poorly constructed arbitrary experiments that say very little about the competency of either model.

The article reads like thin, auto-generated AI clickbait for nerd sniping or shilling a model.

Take a look at the lead:

DeepSeek V4 Pro wins this head-to-head by being more exact where it matters: following instructions, matching schemas, and solving edge cases cleanly. GPT-5.5 Pro is still strong, but it gave away points with avoidable deviations.

“where it matters”, “cleanly”, “is still strong”, and vague references instead of stating that DeepSeek yielded more concise results in 3 out of 4 tests.

1 star.

https://news.ycombinator.com/item?id=48437681

I see many comments saying, “AI can’t do X with 80-100% accuracy; therefore our professions are in good hands.”

While I don’t want to sound overly pessimistic, the models are improving at a rapid rate. If asked ~3 years ago where the state of the models are today, it would sound like sci-fi if answered, “the models are creating full MVP apps in ~30 minutes with one prompt.”

The hurdles the models are facing now, like reducing hallucination rates, ensuring compliance, and keeping a clean codebase, do not seem far away from being resolved IMO. Fetching specific information is already partially done with various MCP servers / RAG.

I am, of course, a bit worried about the future of software engineers. If these quirks are resolved, where do their professions fit in the industry? Delegating tasks to the AI model? Unfortunately, this does not require years of expertise, which is a double-edged sword. Reviewing AI’s output? Ask it to explain each line not understood.

I think we will see more waves of larger layoffs, similar to how human computers were replaced by digital computers. To some, doing complex mathematical calculations mentally is a fun task / challenge, but it is ultimately significantly slower and more error-prone than calculating with a computer. In the same way, I think hand-crafting code will be seen as a fun “challenge” and AI will be seen as the “modern-day calculator”.

george_max

I see many comments saying, “AI can’t do X with 80-100% accuracy; therefore our professions are in good hands.” Although I don’t want to sound overly pessimistic, the models are improving at a rapid rate. If asked ~3 years ago where the state of the models are today, it would sound like sci-fi if answered, “the models are creating full MVP apps in ~30 minutes with one prompt.”

The hurdles the models are facing now, like reducing hallucination rates, ensuring compliance, and keeping a clean codebase, do not seem far away from being resolved in my opinion. Fetching specific information is already partially done with various MCP servers / RAG.

Of course, I am a bit worried about the future of software engineers. If these quirks are resolved, where do their professions fit in the industry? Delegating tasks to the AI model? Unfortunately, this does not require years of expertise, which is a double-edged sword. Reviewing AI’s output? Ask it to explain each line not understood.

I think we will see more waves of larger layoffs, similar to how human computers were replaced by digital computers. To some, doing complex mathematical calculations mentally is a fun task / challenge, but it is ultimately significantly slower and more error-prone than calculating with a computer. In the same way, I think hand-crafting code will be seen as a fun “challenge” and AI will be seen as the “modern-day calculator”.

Switzerland will have a referendum to cap population…

https://news.ycombinator.com/item?id=48451280

Swiss here and able to vote.

In fact, just posted my voting letter today, before taking a 1h bike ride through the biggest city in Switzerland, having lots of space and freedom biking around in our beautiful city.

When taking the train to my parents’ house, I pass several farms and landly smaller cities. A lot of free space in between those, train mostly has spare seats, depending on rush hour timings. There usually are several big commercials on private farmer land stating “NO to 10 Million Population”, prompting people to vote YES on the SVP/UDC initiative.

How’s Linear so fast? A technical breakdown #

https://news.ycombinator.com/item?id=48438545

We use Linear at work. I’m definitely in the minority, but I really struggle with the UX. I also wouldn’t call it fast. Sure the page technically loads reasonably quickly, but half the time I see numbers updating on the page with no visual indicator that data loading is still happening.

HoyaSaxa

We use Linear at work. I’m definitely in the minority, but I really struggle with the UX. I also wouldn’t call it fast. Sure the page technically loads reasonably quickly, but half the time I see numbers updating on the page with no visual indicator that data loading is still happening.

https://news.ycombinator.com/item?id=48445262

Your theory about cable television is fascinating. I never really watched TV growing up, but every time I visit my parents now that they’re retired, one of them always has a 24 hour news going and it’s just non stop “you should be afraid” and “you should be angry” told to you by pretty faces smiling the whole time.

Social media is totally that today too. I quit facebook in 2016 and reddit in 2023 over similar fears. Back then I said facebook was bad for my mental health, and I quit reddit when they made it harder for me to prevent what I called amygdala-bait. But it’s totally the same thing.

These days I love to watch nuanced explanations on youtube of complex issues, but youtube’s algorithm desperately wants to feed me stuff like How Money Works and other channels where it’s dressed up as nuanced explanations of the world, but every single episode is how X is screwing you over or how the Y is going to blow up the economy any second now.

malfist

Your theory about cable television is fascinating. I never really watched TV growing up, but every time I visit my parents now that they’re retired, one of them always has a 24 hour news going and it’s just non stop “you should be afraid” and “you should be angry” told to you by pretty faces smiling the whole time.

Social media is totally that today too. I quit facebook in 2016 and reddit in 2023 over similar fears. Back then I said facebook was bad for my mental health, and I quit reddit when they made it harder for me to prevent what I called amygdala-bait. But it’s totally the same thing.

These days I love to watch nuanced explanations on youtube of complex issues, but youtube’s algorithm desperately wants to feed me stuff like How Money Works and other channels where it’s dressed up as nuanced explanations of the world, but every single episode is how X is screwing you over or how the Y is going to blow up the economy any second now.

https://news.ycombinator.com/item?id=48438279

“ No part of the prose was machine-generated. You will not find machine-written prose on this blog. I consider it deeply disrespectful.”

<3

arthurofbabylon

“No part of the prose was machine-generated. You will not find machine-written prose on this blog. I consider it deeply disrespectful.”

❤️

https://news.ycombinator.com/item?id=48438679

I would always hear about how Linear was fast, but after actually working daily with it, I’ve lost enthusiasm. Search is quite slow, the UI is often clunky (looks good though), “Pulse” is a torrent of noise even at small scale, and I have trouble finding things I need and resort to adding everything to favorites.

Early days’ Trello was the best project tracking experience by far.

ricardobeat

I often hear people say Linear is fast, but after using it every day, the enthusiasm fades. Searching is quite slow, and the interface often lags (although it looks nice), "Pulse" is a bunch of noise even on a small scale, I find it hard to find what I need and end up adding everything to my favorites. Early Trello was by far the best project tracking experience.

https://news.ycombinator.com/item?id=48437896

Whole blog post is basically: Make a mutation in the clientside, assume it worked, and save in the background.

syspec

The whole blog post is basically: Make a mutation on the client side, assume it worked, and save in the background.

https://news.ycombinator.com/item?id=48434590

My career path is surprisingly similar to the author’s. Weirdly enough, what he takes as the first pillar to fall is the one I see most undamaged currently.

LLMs routinely fail at our business specifics: Local tax regulations, particularities of the accounting process, specifics of our ledger implementations. They’re great at refactoring, translating between languages, tracing bugs on existing code even, but there is always many things subtly wrong iterating and expanding our domain.

This might be because the companies I worked for happen to be tackling complex domains precisely for moat-building reasons. They stay in business explicitly because there’s not a book out there you can read to build a clone, the knowhow stays inside.

Also, a fintech whose managers recommend speeding up design docs with AI sounds way too careless to be in the money handling business. It’s way, way too easy to end up with millions incorrectly allocated, particularly if you deal with high volumes of small transactions. These bugs are always a bitch to deal with because correcting the logic is just step one, you then have to correct all the wrongly calculated data in immutable DBs, move around the red tape and client comms, and your fix is bound to become a gotcha that new features and observability have to take into account (“remember that there’s a bump in the data in february 2 because we had incident X”.)

torben-friis

My career path is surprisingly similar to the author’s. Weirdly enough, what he takes as the first pillar to fall is the one I see most undamaged currently.

LLMs routinely fail at our business specifics: Local tax regulations, particularities of the accounting process, specifics of our ledger implementations. They’re great at refactoring, translating between languages, tracing bugs on existing code even, but there is always many things subtly wrong iterating and expanding our domain.

This might be because the companies I worked for happen to be tackling complex domains precisely for moat-building reasons. They stay in business explicitly because there’s not a book out there you can read to build a clone, the knowhow stays inside.

Also, a fintech whose managers recommend speeding up design docs with AI sounds way too careless to be in the money handling business. It’s way, way too easy to end up with millions incorrectly allocated, particularly if you deal with high volumes of small transactions. These bugs are always a bitch to deal with because correcting the logic is just step one, you then have to correct all the wrongly calculated data in immutable DBs, move around the red tape and client comms, and your fix is bound to become a gotcha that new features and observability have to take into account (“remember that there’s a bump in the data in february 2 because we had incident X”.)

xAI is looking more like a datacentre REIT than a … #

https://news.ycombinator.com/item?id=48448520

And Google is a major shareholder in SpaceX, so they certainly have incentive to juice the valuation of the IPO.

Google owns 5-6% of the shares of SpaceX. SpaceX is seeking a valuation of $1.77T which means Google’s shares would be worth $88.5B-$106.2B. I’m not a skeptic of AI/LLMs but this makes me deeply suspicious of these circular deals. What happens when the music stops?

TSiege

Google is a major shareholder in SpaceX, so they certainly have incentive to juice the valuation of the IPO.

Google owns 5-6% of the shares of SpaceX. SpaceX is seeking a valuation of $1.77T which means Google’s shares would be worth $88.5B-$106.2B. I’m not a skeptic of AI/LLMs but this makes me deeply suspicious of these circular deals. What happens when the music stops?

AI is slowing down #

https://news.ycombinator.com/item?id=48449250

undeniable, massive productivity gains.

Take any stock index, remove AI stocks, what do you see? That’s right! Nothing…

So where is all the productivity going? Where is the value? Where are the massive unemployment stats or the millions of new startups making big $$$?

toasty228

Undeniable, massive productivity gains.

Take any stock index, remove AI stocks, what do you see? That’s right! Nothing…

So where is all the productivity going? Where is the value? Where are the massive unemployment stats or the millions of new startups making big $$$?

Pentagon raised threat of Israeli spying on U.S. t… #

https://news.ycombinator.com/item?id=48430768

I don’t like how Americans dismiss their own agency in this all. This is not being forced on us by Israel. A huge and underdiscussed reason why the US and Israel have the relationship they do is because of the roughly 100m American Evangelicals that want it that way. That’s also why the GOP is seen as the party more favorable to Israel (not that the Democrats are necessarily adversarial, but they clearly aren’t as agreeable as a whole).

slg

I don't like how Americans ignore their own agency in all of this. This is not being imposed on us by Israel. A huge and under-discussed reason why the US and Israel have the relationship they do is because of the roughly 100 million American Evangelicals who want it that way. That's also why the GOP is seen as the party more favorable to Israel (not that the Democrats are necessarily adversarial, but they clearly aren't as agreeable as a whole).