Hacker NewsTuesday, October 6

Some ask why anyone would remove small local models. Most say auto-installing ~12-25GB without an easy opt-out is anti-user, especially on 256GB Macs. Many don't want background AI and argue owners should decide what runs on hardware they paid for.

Many compare needing debloat scripts on macOS to Windows tools like win11debloat and O&O ShutUp10. They see Apple following Microsoft toward enshittification, blame forced AI after Apple fell behind, and cite immutable system files that block removal.

Some users report moving to Linux (KDE, GNOME, NixOS) over lockdown, consent prompts and bloat; others defend Apple's ecosystem convenience. Small base SSDs and costly upgrades worsen the issue. Commenters also share tips for deleting Xcode simulators via simctl.

13.3M gallons (~40 acre-feet) is tiny next to one average Nebraska farm (~1,200 acre-feet/yr). Debate follows on whether food justifies farm water, given corn ethanol, alfalfa and livestock feed, and whether evaporated data center water is truly comparable.

Many ask why companies hide figures if they're low. Suggested reasons include trade secrets, revealing compute capacity, stock market signals and routine LLC structures. Others note Google already publishes per-site water data, and skeptics distrust disclosures anyway.

Some see the backlash as broader AI anxiety rather than concern about water. Others cite real local harms: higher power bills, new gas plants, noise, large tax refunds, few jobs, NDAs that sidestep review, and the unprecedented scale of new AI facilities.

Many argue Florida's statute requires a communication "viewed by another person," which a private chatbot diary isn't, and that intent matters. Others note Anthropic's ToS permits human review, making it "transmitted." Most expect a good lawyer to get the charge dropped.

Some sympathize with Anthropic given lawsuits against OpenAI for failing to report a shooter. Critics counter that companies created this dilemma by reading chats at all, and call it mass surveillance. Users should treat chatbots as Big Tech, not a confidant.

Suggestions range from pooling money for H200s to cheap old GPUs running small uncensored models, DGX Spark or Mac Studio setups, or non-US open-weight hosts. Skeptics cite cost, trust, and maintenance, and argue this is a legal and government problem, not a technical one.

Users criticize Ceramic's ToS banning storage, caching, or sharing of output, which blocks agent transcripts and long sessions. Some call it hypocritical given scraped data and doubt enforceability; others suggest weighing intent or simply ignoring terms.

Debate over why Cloudflare resells search APIs: defenders cite unified billing, failover, and single-vendor procurement; critics see rent-seeking, hypocrisy in selling bot protection and crawlers, and worry about pay-to-crawl and "verified" bots controlling web access.

Users compare Gemini Flash Lite 2.5's free Google searches, Kagi, Serper, Perplexity, Linkup, Exa, and self-hosted Firecrawl. Pricing varies widely; testers found Ceramic's results poor and stale, and noted ZDR claims conflict with provider docs.

Many feel forced to give data to untrustworthy firms. Some gave up after futile degoogling efforts. Others say data may be misused by future regimes and urge data minimization, lying on forms or rotating phone numbers. AI medical transcription errors were also raised.

Commenters want hefty fines, jail time or payouts to victims. They note GDPR retention rules are rarely enforced and that governments leak too (Denmark, Berlin, France, Poland). A Polish medical SaaS tested with real patient data and faced no visible penalty.

Users debated what leaked. CPR numbers encode birthdate and sex, and protected addresses were reportedly excluded. Many say CPR should be a username, with MitID handling authentication, though MitID's family-based recovery questions are now weak. Swedes note similar data is public there anyway.

Commenters call Google's quota on non-Samsung OEMs shipping GrapheneOS Microsoft-level anticompetitive, doubt regulators will act, cite EU's past fine, contrast it with Apple's never-open model, and debate whether Motorola will actually ship devices with GrapheneOS preinstalled.

The Pixel 11 has MTE in hardware but not enabled in firmware. GrapheneOS won't support it unless Google ships MTE. Some criticize GrapheneOS for guessing Google's motives (cost-cutting), others debate how much MTE matters and whether Motorola should be the focus.

Many see the Pixel 11 as a minimal, cost-cut update after the RAM price spike and plan to keep older Pixels or buy a used 10. Others note Pixel 8 and 5a motherboard failures and poor warranty handling, and argue phones are now "solved," making frequent upgrades pointless.

Users miss drag-and-drop palettes, property grids, double-click event handlers, F1 context help and edit-and-continue debugging, wondering why modern tooling lost these. Delphi, WinForms and Anvil are cited as descendants or modern equivalents.

Skeptics say VB6 was messy beyond simple GUIs, palettes don't scale and absolute layouts break with resizable windows and responsive design. Others counter that basic desktop apps need little more, and VB6/Delphi were peak IDEs.

Commenters note visual glitches like crooked buttons, missing bevels and wrong border corners, plus selection and event-list bugs and LLM-sounding labels. The author's huge AI-assisted output impresses some but raises slop concerns; the author welcomes fixes.

Amazon and others fly in German or Belgian technicians for skilled automation work while US workers get manual jobs. Commenters argue over German salary costs, critique short-term cost-cutting and lack of apprenticeships, and dispute claims that Germany's economy is thriving.

Most investors are American, raising worries the company will relocate to the US like Heart Aerospace. Others call this a purity test, noting money is global. Some cite risk-averse European capital and low German stock ownership as reasons startups rely on US VCs.

Many note RobCo is an evil corporation from Fallout, joking about Pip-Boys, Robert House, Aperture Science and Umbrella. Comparisons are made to Palantir, Anduril, Stark, Mordor Intelligence and the "Torment Nexus" phenomenon.

Many argue WebComponents have clumsy APIs: string-only attributes, attribute/property syncing boilerplate, Shadow DOM CSS pains, no declarative state. Most need Lit or Stencil to be usable. A few defend them as fine, fast, and universal for simple sites or leaf components.

Built-in elements like datalist, date/time pickers, dialog, and multiselect are often buggy, inconsistent across browsers, unstyleable, or lack features like search and ranges. Designers and users demand more, so developers roll their own or use libraries despite accessibility trade-offs.

React and similar tools emerged to patch a painful platform, like footpaths forming where sidewalks are missing. Commenters debate whether React's dominance stems from real merit (state management, ecosystem, DX) or hype and docs, and note browsers later standardize what developers hack together.

An off-topic subthread dominated. Some cited the BND's 80-95% lab leak assessment, WIV gain-of-function history and China's obstruction. Others pointed to the Rootclaim debate and a scientific consensus favoring zoonosis. Many called the question unresolved and divisive.

Commenters asked why the exposure didn't trigger immediate antibiotics and containment, and debated glass tubes and Russian safety culture. Others called the "broken tube" account a rumor, citing a fieldwork theory, government denials and past Russian cover-ups.

Some questioned the value of such labs, noting the post-COVID risk debate never happened due to grant incentives and suppressed lab-leak talk. Others said plague occurs naturally and needs vaccine research, and pointed to wet markets and past lab accidents like smallpox.