HN Brief: 2026-05-22
Today’s HN was split between AI backlash and hardware optimism. The plagiarism and slop-grenade threads turned into a reckoning over whether AI is just theft at scale or a genuinely useful tool buried under hype, while Google caught flak from two directions—Antigravity IDE users mourning yet another abandoned product, and everyone watching its ad business pivot into AI-generated search results. Meanwhile, the Flipper One announcement kicked off a fierce form-factor war (QWERTY vs. pocketability) and the Trinity test photos pulled the crowd into a grim meditation on nuclear physics and Christopher Nolan’s botched visuals. The mood wasn’t hype; it was people arguing about what’s worth building and what’s already broken.
Threads most worth clicking: “Flipper One – we need your help” for the keyboard-versus-compact civil war that cuts to the heart of what a Linux handheld should be. “AI is just unauthorised plagiarism at a bigger scale” because it forced the crowd to admit that scale and centralization change the moral calculus even if the mechanism isn’t new. “Waymo pauses Atlanta” gets at a rare honest admission about autonomy’s physical limits—water detection is a sensing and training-data problem, not a software one. “Lost Images from the 1945 Trinity Nuclear Test Restored” for the genuinely unsettling restored photos and the unanimous dunking on *Oppenheimer*’s underwhelming blast. And “Earth is now heating up twice as fast” for the shipping-sulfur twist and the cold math on how fast we’re burning through the 1.5°C guardrail.
Flipper One – we need your help [comments]
1132 points · 441 comments · blog.flipper.net · 20h ago
The Flipper team announced the Flipper One, a Linux cyberdeck built around an open-hardware philosophy with a co-processor architecture, aiming to be the most blob-free ARM computer possible — completely separate from the Flipper Zero. A huge chunk of the thread immediately turned into a form-factor argument: one camp wanted a built-in QWERTY keyboard and x86 power (pointing to GPD Pocket devices as the obvious alternative), while the other side pushed back hard, arguing the whole point is a pocketable, rugged tool where you bring your own peripherals and the compact size is a feature, not a bug. The other major fault line was scope creep — plenty of people called the project overambitious, predicting it won’t ship before 2030 (if ever), especially given the team is building their own OS profiles and planning a custom LLM; but defenders pointed to the Zero’s early doubters and said the passion and open-source commitment is exactly what makes it worth betting on. A few technical tangents dug into the co-processor architecture being actually standard (every laptop does it with the BIOS/Secure Enclave), and some skepticism about the 8 GB RAM ceiling given the local SDR and AI ambitions. Underneath it all, the real excitement revolved around the promise of full mainline kernel support and zero binary blobs — that’s the differentiator that even the keyboard critics conceded they couldn’t get anywhere else.
Project Hail Mary – Stellar Navigation Chart [comments]
837 points · 183 comments · valhovey.github.io · 15h ago
The linked article wasn’t available to this summarizer; from the discussion, it’s an interactive stellar navigation chart built around the physics of *Project Hail Mary*. The thread immediately spins into a sprawling mix of book and movie appreciation—people stacking their favorite popcorn sci-fi (Bobiverse, Expeditionary Force, Becky Chambers) and arguing whether those belong in the same bucket—alongside a giddy deep-dive into the actual astronomy: fans debate whether the Petrova lines curve correctly, point out that the planets’ orbital radii are off by a factor of 400, and measure pixels to prove Tau Ceti appears 600 times too close. There’s a persistent side argument about whether the whole thing looks AI-generated (some say “same font as every AI viz,” others counter that polish is now easier for everyone), plus a tangent on gravity well diagrams that turns into a full vector field physics lesson. A few people pivot to *The Expanse* and how even that show fudges engagement ranges, and the thread wraps with a lot of genuine gratitude and links to time-dilation visualizations.
AI is just unauthorised plagiarism at a bigger scale [comments]
778 points · 677 comments · axelk.ee · 18h ago
The article is a first-person complaint from a tutorial author who found ChatGPT-using competitors copying his work, ranking above him in Google, and even leaving his original links intact—a speed-of-light version of something that’s been happening for decades. The thread split hard: a large faction dismissed the post as a naïve rant, pointing out that content theft and SEO gaming predate LLMs by twenty years and that the author’s specific evidence (identical link text, full article rewrites) sounds like old-school manual plagiarism, not something ChatGPT uniquely enables. But the counterargument that carried real weight was about scale and morals—yes, plagiarism existed before, but giving anyone with a prompt the ability to wholesale repackage another person’s expertise into a saleable product, with zero expertise or effort on the plagiarist’s part, changes the game. The deeper argument that got traction was that AI companies aren’t democratizing knowledge; they vacuumed up the entire public commons into proprietary models owned by a few megacorps, then sell access back to us, which is a concentration of power, not liberation, and the “information wants to be free” crowd has to reckon with who actually owns the free stuff now.
Google's Antigravity bait and switch [comments]
654 points · 298 comments · www.0xsid.com · 18h ago
The article is a detailed complaint about Google’s Antigravity IDE being silently replaced by a chatbot-style 2.0 version via background update, nuking the old interface and breaking the author’s production workflow, with a painful purge-and-reinstall required to get the legacy tool back. The thread largely ignored the technical gripe and instead turned into a broader post-mortem on Google’s AI downfall—most commenters argue Google squandered its years-long research lead because it can’t stop shooting itself (and its customers) in the foot with half-baked product migrations and a refusal to cannibalize search revenue. A loud faction insists the real problem is that Gemini models simply aren’t competitive for coding compared to OpenAI and Claude, though some push back that Google still leads in open-weight models like Gemma and in image generation. A significant side tangent dove into astroturfing accusations, with several people claiming they’ve been paid to shill for various AI labs on HN, leading to meta-discussion about how to identify and report such behavior. The consensus is that Google’s enterprise tools have been declining for over a decade, and this Antigravity mess is just the latest example of why no one trusts them with anything serious.
We're testing new ad formats in Search and expanding our Direct Offers pilot [comments]
599 points · 535 comments · blog.google · 22h ago
Google is embedding Gemini into search to roll out new AI-generated ad formats—think conversational product guidance and sponsored entries inside AI Mode—along with expanding a Direct Offers pilot that bundles promotions with checkout integration for travel and retail. The thread immediately zeroes in on the obvious: Google is an ad company, and with search ads accounting for 57% of Alphabet’s revenue, this was never not going to happen; the real question is how transparent the ads will be when the AI can weave them seamlessly into answers. A big split forms over whether Google should have waited for OpenAI to blink first, since the first mover on AI search ads takes a reputational hit, but others argue the public barely notices ads anyway and the real pressure is from advertisers wanting access to this format. The deeper concern—that the AI’s answers will be influenced by who pays, not just what’s correct—gets called the elephant everyone *is* talking about, with comparisons to existing map routing rumors and the impossibility of inspecting a synthesized answer like you can a page of results. Some push back that local AI won’t help if the training data is already SEO-optimized, while a chunk of the thread devolves into arguing whether stock prices reflect fundamentals or just vibes, and whether 30% ad-blocker adoption is a massive threat or a rounding error.
Throwing AI-generated walls of text into conversations [comments]
586 points · 346 comments · noslopgrenade.com · 22h ago
The linked article isn’t a blog post per se — it’s a mini-manifesto against pasting massive AI-generated responses into chat or email when a one-sentence human answer would do, framed around the Redis-versus-Memcached question and coining the term “slop grenade.” The HN crowd mostly agreed with the premise, but immediately turned the weapon back on the author: a top reaction was that the blog itself should have been a comment — it’s “another blog post that should have been a comment,” and the site was called “pure slop designed in a pig lab for HN trough.” A big split emerged over whether long, context-heavy messages in Slack are ever appropriate: some argued that providing full context is respectful and necessary, especially for nuanced work questions, while others countered that walls of text are hostile, declare authority prematurely, and kill the back-and-forth that makes conversation useful. There was also a self-referential meta discussion about using AI to summarize AI-generated slop, and a deeper critique that the example (Redis vs. Memcached) was too narrow and tech-bubble, making the whole thing a performative insider shaming tool rather than a broadly useful social norm.
Seattle Shield, an intelligence-sharing network operated by the Seattle police [comments]
455 points · 185 comments · prismreports.org · 14h ago
The article details Seattle Shield, an intelligence-sharing network run by the Seattle police that includes Facebook, Amazon, ICE, and a host of other private and law enforcement entities, with reports focused almost exclusively on protests. HN immediately zeroed in on the roster's weirdest members—the Church of Scientology pops up, which sent the thread down a rabbit hole on L. Ron Hubbard's history and the group's long track record of infiltrating government agencies. A significant chunk of commenters dismissed this as a nothingburger, arguing that police departments have always coordinated with local businesses and that the evidence shows no actual abuse of power, just an email list. Others pushed back hard, saying the very existence of a permanent, secretive public-private fusion center modeled on post-9/11 NYPD Shield is exactly the kind of normalized surveillance that should alarm people, especially after the BlueLeaks breach exposed the whole membership list. The debate largely split between those who see a mundane mailing list and those who see a panopticon in the making, with the Scientology tangent nearly hijacking the whole conversation.
Indexing a year of video locally on a 2021 MacBook with Gemma4-31B (50GB swap) [comments]
368 points · 107 comments · github.com · 17h ago
The article walks through a local pipeline that extracts metadata, transcribes speech, detects faces, and runs a vision model across a year's worth of raw safari footage on a 2021 MacBook M1 Max with 64GB RAM, using Gemma4-31B and heavy swap. HN was split: some were genuinely impressed by how far local models have come, swapping experiences about M1 vs M5 performance and praising the sheer integration work, while a long, heated subthread tore into the blog post itself for being painfully obvious AI-generated slop — bolded bullet points, formulaic phrasing, the works. The author took the criticism well, even sharing a link to a "humanizer" tool someone recommended, but others argued that running it through another AI pass to hide the stink doesn't fix the lack of substance, and that the real substance (the pipeline) is what saves the post. A smaller tangent caught the author accidentally exposing a `~/.claude/skills/` localhost path in the post — classic "I put something on my local server and shared the link" moment.
Python 3.15: features that didn't make the headlines [comments]
367 points · 182 comments · blog.changs.co.uk · 20h ago
Python 3.15's feature freeze brings some genuinely useful quality-of-life improvements—TaskGroup now has a clean `.cancel()` method, context managers work properly with async functions and generators, there's a whole new thread-safe iterator API, `Counter` supports XOR, and `json.loads` can return immutable data structures via `frozendict` and the new `array_hook`. But the thread immediately veered into a much bigger fight about whether Python has a future outside machine learning, with multiple people saying they're actively deleting tens of thousands of lines of Python and moving to Go or Rust. The argument splits into two camps: one side says Python's slow runtime, weak typing story, and GIL make it a liability now that AI coding tools write decent Go/Rust but terrible Python, while the other side pushes back hard that Python's indentation and REPL are still unbeatable for rapid prototyping and that syntax complaints are overblown. A surprising number of people claim AI tools actually produce *better* code in typed, compiled languages because hallucinations are easier to catch, and that Python's vast but inconsistent training data (Python 2 vs 3, no standards) is actually a liability for code generation quality. The whole thing feels less like a conversation about 3.15's features and more like a reckoning over whether Python earned its dominance or just got lucky with the AI boom.
Shunning AI is the human choice [comments]
357 points · 503 comments · www.thehandbasket.co · 18h ago
The linked article is an op-ed arguing that rejecting AI is a legitimate personal and cultural choice, not a technophobic holdout. The thread immediately split into two opposing camps: one side insisting AI is an inevitable technological force that critics must accept, the other pushing back that inevitability is a rhetorical cudgel, not a fact. Several people invoked historical parallels, from 1980s database resistance to crypto and NFTs, to argue that hype cycles can collapse; others countered that unlike fads, LLMs are a genuinely useful technology that people are already running locally, not just in corporate data centers. A recurring pointed exchange compared disliking AI to being stuck in traffic—ubiquitous but not welcomed—while a deeper strain of the conversation questioned whether the real problem is the marketing and power structures around AI (the “slop,” the job replacement rhetoric) rather than the technology itself, with one person neatly distinguishing between “AI as useful tool” and “AI as cultural envelope.”
Lost Images from the 1945 Trinity Nuclear Test Restored [comments]
338 points · 101 comments · spectrum.ieee.org · 20h ago
The article showcases restored, high-speed photographs from the Trinity test, offering startlingly clear views of the first milliseconds of the nuclear age. The thread immediately split on whether the actual earliest frames are still classified for their detonator tech, with some insisting fission weapons are trivial now and others countering that precision implosion details remain restricted. A major, near-unanimous side tangent tore into Christopher Nolan's *Oppenheimer* for botching the blast with underwhelming practical effects that looked like ordinary chemical explosives, missing the "unearthly cosmic horror" that made witnesses quote scripture. The discussion then spiraled into grim speculation about the last human on Earth and whether the end will be a quick fireball or slow starvation, with one camp arguing the real horror is how abstract math and centuries of theory produced this violence—and that we're nowhere near responsible enough to wield that kind of energy.
Waymo pauses Atlanta service as its robotaxis keep driving into floods [comments]
304 points · 377 comments · techcrunch.com · 15h ago
Waymo has paused service in four cities after its robotaxis kept driving into flooded streets, with the company admitting its fix isn't fully baked yet and that flooding in Atlanta happened before weather alerts could trigger a shutdown. The thread mostly brushed past the engineering question and went straight to arguing about whether autonomy should even try to handle every condition — with one side saying you just accept weather limits and shut down when it’s bad, and the other side claiming full self-driving will eventually handle hurricanes through coordinated evacuation and bumper-to-bumper platooning. A lot of the pushback came from people saying the real bottleneck isn't sensors but physics: roads have finite capacity, and even perfect reaction times can’t double throughput without trains. The technical deep dive centered on whether water detection is a sensing problem or a decision-model regression issue, with someone pointing out that point clouds can reveal standing water if you don’t filter the ground plane out, but it’s likely missing from the training data.
BBEdit 16 [comments]
300 points · 91 comments · www.barebones.com · 13h ago
BBEdit 16 shipped with over a hundred changes—performance tweaks, AI chat improvements, text search inside images, and overdue vi keybindings for basic navigation. The thread largely treated this as a celebratory check-in on a beloved Mac editor, with old-timers reminiscing about the Classic Mac OS days and praising Bare Bones for sticking to a sane paid-upgrade model ($30 for recent licensees) instead of subscriptions. A chunk of the discussion veered into a pricing-model debate: some argued that fixed-price perpetual licenses are the only honest way to sell software, while others countered that ongoing security and macOS compatibility work justifies subscription revenue, though most agreed BBEdit’s approach is a rare bright spot. The vi emulation split the room—a few shrugged it off as a “meh” major-version feature, but longtime Vim converts called it a godsend. A separate tangent dug into Yojimbo, another Bare Bones product that hasn’t seen real updates since 2020, sparking a nuanced exchange about when “finished” software is acceptable versus when it just feels abandoned.
Show HN: Freenet, a peer-to-peer platform for decentralized apps [comments]
272 points · 170 comments · freenet.org · 17h ago
The linked article announces the relaunch of Freenet, a peer-to-peer platform for decentralized apps, rebuilt from scratch by original creator Ian Clarke with a unique key-value store where WebAssembly contracts define commutative merge operations for state synchronization. HN's technical crowd immediately recognized this as closely related to CRDTs and CvRDTs, with the submission's author confirming that Freenet lets each contract define its own merge semantics in WASM, essentially making each contract a custom CmRDT. A significant thread dug into the system's limits: unlike blockchains, Freenet doesn't guarantee global consensus or solve double-spend problems, making it unsuitable for cryptocurrencies but fine for group chat and social networks where temporary inconsistency is acceptable. Several commenters challenged the project's approach to sybil resistance and reputation, arguing that the current reliance on "ghost keys" funded by donations to the Freenet foundation is inherently centralized, while the author acknowledged this is just one bootstrap option and pointed to decades-old web-of-trust experiments on the original Freenet as proof-of-concept for decentralized governance.
News outlets are limiting the Internet Archive’s access to their journalism [comments]
267 points · 92 comments · www.niemanlab.org · 15h ago
More than 340 local news outlets are now blocking the Internet Archive’s crawlers, worried that AI companies will scrape their journalism out of the Wayback Machine for training data. The HN crowd is sharply split: some see the move as shortsighted panic that will obliterate the historical record, pointing out that a time-delayed archive (say, after a week) would preserve the public good without undercutting paywalls or licensing leverage. Others agree with the publishers’ position, arguing that if the Internet Archive serves up current articles for free, it directly competes with the original source and undermines any chance to negotiate with AI companies—and they push for a micropayments model where OpenAI or Google pays a few cents per read. A few commenters go deeper, noting that the real fight is over copyright and fair use, with one judge’s whim deciding whether archival access counts as fair or as piracy, while another thread devolves into a squabble about Cloudflare captchas and the impossibility of anonymous micropayments. Meanwhile, several people share horror stories of articles being memory-holed entirely—even the Wayback Machine showing nothing—and cite the need for decentralized, client-side archiving tools that don’t depend on a single nonprofit’s goodwill.
Blog ran on Ubuntu 16.04 for 10 years. I migrated it to FreeBSD [comments]
261 points · 142 comments · crocidb.com · 13h ago
The article is a detailed walkthrough of migrating a blog from an ancient Ubuntu 16.04 VPS to a cheaper, more powerful FreeBSD server using Jails, ZFS, and Bastille, with benchmark results showing massive performance gains (though mostly from newer hardware). The comments largely ignored the FreeBSD specifics and instead turned into a sprawling debate about long-term support Linux distros—people argued over whether Alma/Rocky or Debian LTS are better for low-maintenance servers, with some insisting that upgrading every few years is trivial while others countered that even simple upgrades have broken their machines or that they just want zero maintenance. A vocal contingent pitched NixOS as the real answer, praising its immutable rollback and declarative config, though detractors complained about flake verbosity and nondeterministic builds eating hours. The thread also wandered into a brief side debate on formal education vs self-teaching, sparked by one commenter’s appreciation for hands-on experimentation.
US employers spend more than $1.5B a year to fight labor unions, report finds [comments]
242 points · 201 comments · www.theguardian.com · 18h ago
A new report from the Economic Policy Institute found that US employers spend over $1.5 billion a year on union-avoidance consultants and law firms, with Amazon alone dropping $26.6 million in 2025. The thread immediately latched onto that number as pocket change — several people pointed out that $1.5B is a rounding error compared to the estimated $15B+ in annual wage theft just from minimum wage violations, so the spending is a rational investment that pays for itself many times over. A large chunk of the discussion then veered into a familiar and fairly split debate about whether unions themselves inevitably become corrupt power centers, with people citing the longshoremen's union leader who went on TV flashing a Rolex and threatening to cripple the economy, while others countered that this is a tired anti-union trope that doesn't hold up when you look at European unions or worker cooperatives. There was also a meatier tangent arguing that unions are just corporations that sell their members' labor collectively, and if you're fine with Bob's Heavy Manufacturing Concern bargaining as a bloc, you should be fine with workers doing the same — though this got pushback from people who pointed out the difference between voluntary exclusive contracts and being forced to join a union to keep your job.
A Bipartisan Amendment Would End Police License Plate Tracking Nationwide [comments]
238 points · 85 comments · www.wired.com · 19h ago
The article covers a proposed bipartisan amendment that would block any recipient of federal highway funding from using automated license plate readers for anything other than tolling, effectively killing most police ALPR programs nationwide. The Hacker News crowd was immediately suspicious of federal motives, arguing that the feds want to shut down state and local surveillance to protect their own pet projects—like the DEA’s I-95 corridor—while public anger builds against the “fund everything with strings attached” paradigm. Others pointed out that even if this passes, private car manufacturers could still collect and share plate data, though several commenters noted that many states already ban private collection, and one linked to a map of state laws. A major split emerged over speed and red-light cameras: some argued that banning ALPRs would also kill those proven safety tools, while others shot back that those cameras are themselves intrusive, catch ordinary people making small mistakes, and create a permanent surveillance record that inevitably gets abused—as in the Texas deputy who tracked an abortion patient. The deeper consensus was that even limited ALPR use suffers from unstoppable mission creep, and without warrant requirements and strict retention limits, any database is a loaded gun for future abuse.
Show HN: I Dedicated 4 Years to Mastering Offline Password Cracking
226 points · 46 comments · news.ycombinator.com · 19h ago
The linked article wasn't available to this summarizer; from the discussion, it's a self-published book on offline password cracking with Hashcat, written by an 18-year-old who spent four years on it. The thread is overwhelmingly congratulatory about the achievement, but several commenters with domain expertise dug into the Amazon sample and pointed out imprecise phrasing and grammar issues — one nitpicked the claim that deriving input from a hash is "literally impossible," given the whole book is about guessing inputs. A warning popped up that hosting the video on Google Drive and selling on Amazon could get his accounts banned, since the platforms might flag it as cracking material despite being about ethical pentesting. A smaller split emerged: someone argued the audience for offline password cracking is mostly criminals, which was pushed back by offensive security pros who use these tools daily for authorized work.
Samsung chip workers will get an average $340k bonus as AI profits soar [comments]
213 points · 166 comments · qz.com · 6h ago
The article reports that Samsung chip workers are in line for average bonuses of around $340,000 after a union deal tied 10.5% of operating profit to payouts, ending a strike that threatened billions in losses. The thread immediately splits into two camps: one group celebrates this as proof that unions and collective action actually work to capture AI windfalls for labor, while another dives into a defensive comparison with U.S. tech workers, arguing that developers earning FAANG money and equity shouldn't complain about their own lack of union power. A significant portion of the comments veers into a sprawling debate about whether American 401(k)s and stock ownership are a form of public ownership of the means of production, with one side arguing the system already distributes wealth broadly and the other countering with data that the top 10% own 90% of that wealth. There's also pushback on the idea that this bonus is pure gravy, with some pointing out that Samsung employees had their payouts artificially capped at half their salary until this strike broke that ceiling, and that SK Hynix workers are actually getting even bigger cash bonuses. The overall vibe is less about Samsung specifically and more about using this story as a flashpoint for a larger argument on whether tech workers in the U.S. are being played by equity compensation and anti-union culture.
Get your passwords out of Bitwarden while you still can [comments]
211 points · 186 comments · www.osnews.com · 17h ago
The article warns that Bitwarden is heading for enshittification under a new CEO, with a price hike, the quiet removal of its "Always free" pledge, and a values rewrite that dropped Inclusion and Transparency for Innovation and Trust. The thread mostly agrees the caution is justified but pushes back hard on the "while you still can" alarmism—most people think export will remain available, pointing out that no major password manager has ever blocked it and that a LastPass-style rug pull would trigger a class-action lawsuit. Several commenters instead use the moment to evangelize alternatives like `pass` (which stores encrypted files in a git repo), self-hosted Vaultwarden, or KeePassXC, though others argue that for non-technical family members, convenience trumps purity and a good hosted service is still better than sticky notes. A recurring side conversation digs into the practical pain of syncing open-format password files across mobile devices without a proprietary sync layer.
Using Kagi Search with Low Vision [comments]
188 points · 59 comments · veroniiiica.com · 12h ago
The article is a personal account from a low-vision user who finds Kagi's paid, ad-free search engine vastly more usable than Google thanks to custom CSS, keyboard shortcuts, and reduced visual clutter. The Hacker News thread mostly agreed that Kagi is a genuinely better experience, with many commenters sharing similar stories about how Google's results pages are an accessibility nightmare for both low-vision and neurodivergent users. However, a vocal minority pushed back hard on privacy grounds, arguing that tying all search history to a paid account with payment information is actually worse than using Google without logging in—a claim that was quickly swatted down by others pointing out Google's shadow profiling and the existence of Kagi's privacy pass and anonymous payment options. There was also an unexpected tangent about the author's own blog design: several readers found the oversized text and line spacing hard to read on mobile, ironically highlighting how accessibility settings that help one person can hurt another.
Uv is fantastic, but its package management UX is a mess [comments]
186 points · 97 comments · www.loopwerk.io · 11h ago
The article argues that while Astral's uv is fast and powerful, its CLI for maintenance tasks like checking outdated packages and upgrading is clunky compared to pnpm or Poetry—no `uv outdated` command, unsafe no-upper-bound version constraints by default, and a verbose `uv lock --upgrade-package` flag repetition. The HN thread pushed back hard: several commenters pointed out that uv intentionally avoids upper bounds because Python’s package ecosystem can’t handle diverging major versions across transitive dependencies the way npm or Rust can, and that semver isn’t reliably followed in Python anyway. Others dismissed the criticism as overblown QoL whining, noting that lockfiles and explicit version pinning solve the upgrade safety issue, or that the `--bounds` flag already exists as a preview. A separate vein of discussion questioned whether uv is genuinely better or just a fast new tool undergoing the same feature creep as its predecessors, with some defending it purely for its dependency resolution speed on gnarly projects.
London Mayor Blocks Palantir [comments]
185 points · 58 comments · www.theguardian.com · 15h ago
Sadiq Khan blocked a £50m Met Police contract with Palantir over procurement violations, saying the force only seriously considered one supplier. The thread largely celebrates the decision, with many arguing Palantir’s ties to US intelligence, ICE, and the Israeli military make it an unacceptable partner for British policing, and that the “land and expand” sales tactic deserves suspicion. Some pushback, though, points out Palantir’s actual technology is just data integration and that the company has delivered real operational improvements, but those arguments get swamped by deeper gripes about US foreign policy and Peter Thiel’s politics. A recurring theme is that this isn’t about technical merit at all — it’s a political signal that London won’t do business with a company that helps bomb children in Gaza, and a broader rejection of the current US regime’s reputation.
Show HN: Rmux – A programmable terminal multiplexer with a Playwright-style SDK [comments]
170 points · 83 comments · github.com · 22h ago
The submission is a Rust rewrite of tmux called Rmux that adds a typed async SDK and native Windows support, positioning itself as a “multiplexer engine” you can build other multiplexers on top of. A chunk of the thread got sidetracked by the project’s AI-generated landing page—some called it a “claudism” with choppy parallax and a pulsing green dot, while others pushed back hard, arguing the code quality matters more than the marketing tools used. The real technical debate was whether Rmux actually improves on tmux or Zellij: the author argued the SDK lets you do automated acceptance tests on TUIs and that its async “wait for text” primitives beat shell-scripting with grep and sleeps, but skeptics wanted a CLI that agent workflows could drive without writing Rust. A correction came in that tmux is written in C, not C++, which the author fixed. One interesting tangent compared the session-pane coupling design to alternatives like abduco and zmx, though the author admitted they kept things tmux-compatible to avoid alienating users.
Chewing gum restores dad's taste and smell years after Covid [comments]
167 points · 93 comments · discover.swns.com · 14h ago
The article reports on a University of Nottingham pilot study where chewing gums with intense, rotating flavors helped restore taste and smell in a man who lost both to COVID years ago. The HN thread immediately split over the details: a bunch of people pointed out that capsaicin “heat” is a pain sensation, not a taste or smell, so the article’s claim that he couldn’t feel spicy food at all suggests COVID can knock out trigeminal nerve responses too—multiple commenters shared their own experiences of losing spiciness, wasabi, and even the nose-burn from mustard, while others retained it. The personal anecdotes took over, with people describing how devastating anosmia is (texture without flavor is “horrifying,” coffee and chocolate permanently dimmed) and trading tips like smell training with essential oils or B-complex supplements for nerve repair. A smaller, testy tangent erupted over vaccine boosters, with one side arguing immunity wanes and the other rolling their eyes at the “still doing this” reflex. The dominant takeaway wasn’t controversy, but hunger: everyone wants to know where to buy this gum.
The memory shortage is causing a repricing of consumer electronics [comments]
162 points · 167 comments · davidoks.blog · 10h ago
The article argues that the explosion of AI demand for HBM memory is starving the supply of LPDDR and DDR memory used in phones and laptops, causing prices to spike and reversing decades of falling consumer electronics costs. The discussion largely accepted the core thesis, with many commenters digging into the mechanics of the memory market and the three-company oligopoly that controls production. A significant split emerged over whether the declining smartphone market is primarily due to this memory crunch or simply because phones have become good enough that people don't feel the need to upgrade—several people pointed out they're still using five-year-old flagships with no real complaints. There was pushback on the idea that cheap phones are disappearing forever, with some noting the massive secondary market for used premium phones could absorb demand, while others countered that the article's focus is on bottom-of-the-market manufacturers in Africa and Asia who rely on repurposed cheap components. A few commenters questioned why Apple, sitting on enormous cash reserves, doesn't just build its own memory fabs—but the consensus was that the multi-year investment horizon and risk of market shifts make even the biggest tech companies reluctant to break the oligopoly's grip.
The IBM-ification of Google? [comments]
160 points · 123 comments · zeroshot.bearblog.dev · 8h ago
The article argues that Google is following IBM’s path to irrelevance, citing a string of failures: GCP nuked a billion-dollar startup’s account with zero human intervention, Search is just repackaged blog content stripped of attribution, and the company kills beloved products like Reader and Inbox so relentlessly that nobody trusts anything new. The HN thread split hard on the “Google kills products” complaint—some pushed back that this is just what healthy risk-taking looks like, pointing to Waymo and Gemini as evidence Google still makes big bets that pay off, while others countered that the quality of a product is no insurance against getting axed, citing hardware that became useless when the cloud service died. A major correction landed on the Reader shutdown: it wasn’t just a casualty of “shiny object” syndrome, it was a deliberate strategic move to funnel users into the doomed Google+ social experiment. The real consensus that emerged wasn’t about product graveyards, but about stagnation—Gmail, a core product, still can’t thread emails in reverse chronological order after decades, while Google pours resources into AI features no one asked for. The comparison to IBM got dismissed as melodramatic doomer bait, with one reminder that Google’s stock was $60 when Reader died and the company’s now worth $4.6 trillion, but another warned that monopolies eventually rot from within when they can subsidize broken business units with ad cash.
Earth is now heating up twice as fast as in previous decades [comments]
156 points · 140 comments · www.newscientist.com · 21h ago
The New Scientist piece reports a new analysis showing Earth’s warming rate has doubled since 2013–14, now at 0.36°C per decade, which could push us past 1.5°C by 2028. The thread immediately split into fatalism—many accepted the breach as already inevitable, citing Berkeley Earth data showing 2023 and 2024 already above 1.5°C—and a brawl over who’s to blame, with the usual per-capita-versus-total-emissions standoff between US and China defenders. Several people pointed to the 2020 shipping sulfur crackdown as the likely driver of the acceleration, noting the aerosol cooling effect was masking warming and its removal means the rate might slow again. A long side-argument erupted about whether AI and crypto energy use matter at all, with one camp saying “fix the generation, not the consumption” and the other insisting we don’t have time to build renewables fast enough to justify new data centers. The boiled-frog metaphor got dissected, too—turns out the original experiment used brainless frogs, which made the analogy either completely wrong or disturbingly apt.
Who wins and who loses in prediction markets? Evidence from Polymarket [comments]
143 points · 139 comments · huggingface.co · 19h ago
The paper analyzes 588 million trades on Polymarket and finds that the top 1% of users capture 76.5% of profits, with successful traders mostly providing liquidity via limit orders while losers take liquidity with market orders. The HN thread immediately got sidetracked into a meta-debate about whether the abstract (which one of the authors posted verbatim in the comments) was AI-generated—people were pointing at the em-dash usage and overly-verbose phrasing as telltale signs, though others pushed back that academic abstracts just read that way and the author was legitimately engaging with questions. A separate line of argument pushed back on the paper's conclusion that insider trading doesn't explain the top winners, with several commenters noting that insiders would naturally use single, short-lived bets and wouldn't show up in monthly persistence metrics. The discussion also split on whether prediction markets are inherently zero-sum (where only market makers and the patient can reliably profit) versus the stock market, which some argued isn't zero-sum because of dividends and capital allocation—though others countered that market cap gains aren't real money you can collectively cash out.
Generated 2026-05-22 08:35 UTC
Generated by Sauron from Hacker News discussions and linked articles.