60 stories
Against M5, Apple claims graphics up to 30 percent faster, AI workloads up to 25 percent faster and a Neural Engine up to 1.9x faster. Memory is 16 GB standard, 32 GB maximum. The quoted bandwidth became the thread's sticking point.
Owners of older machines used it to say they are not upgrading.
The top HN threads: facts from the discussions, both sides, insider voices. Email, Telegram, or RSS.
Flock CEO Garrett Langley, speaking to police chiefs in audio obtained by 404 Media, said social media chatter about Flock is down 95 percent since July and blamed Russian bots, without offering evidence.
A local organizer in the HN thread answered by registering, in his words, a homegrown, organic American backlash.
Microsoft is offering voluntary exits to about 60,000 staff in the US, UK, Canada and India, a few months after an early-retirement round that about 30% of eligible employees took. The company expects fewer takers this time.
In the thread, an ex-Microsoft engineer's reading was that the best people leave first.
The 8.5-inch iPad mini starts at $699 for 128GB, with an OLED display peaking at 1,200 nits and a chip Apple says is 4.5x faster at AI work.
A Canadian commenter put the arithmetic plainly: a 32GB mini cost $379 in 2015, and a 128GB one is $949 today.
The author's favorite explanation is not generosity. Models are now swappable through routers like OpenRouter, and 29 hosts already serve one open DeepSeek model.
The thread's most practical answer came from a team that left for cheaper models over refusals.
Xbox says it has "record-breaking interest in our I.P." and points to upcoming Minecraft and Diablo animated series with Netflix. Gears of War, The Elder Scrolls and Sea of Thieves are reportedly being pitched.
A commenter who works through deal terms said the headline number says almost nothing.
Citing the Wall Street Journal, Blank says Elon Musk personally ordered limits on Starshield for Ukraine. Europeans now ask whether Microsoft, Oracle or Palantir could do the same.
The thread argued about treason, then about whether Europe has any rockets.
The NY Post headline says Google cut down 250K trees illegally. A thread commenter reduced it to 1.5 square miles of planted forest cleared before a legally required environmental assessment.
Finns in the thread said that is exactly why the country is angry.
The Danish Ministry of Digitalisation dropped the Collabora pilot, delivered by Danish vendor cBrain, without announcing it. The paywalled article does not show the ministry's reason in its visible text.
HN supplied its own: a commenter has tried switching every year just to print adhesive labels.
The demos show a malicious PDF opened through iMessage or Signal ending in code execution, followed by an app that roots the phone. Calif says most of the bugs behind them already went to Apple.
The thread split over a different question: is it responsible to show a working takeover before a patch exists, or is the publicity what gets people to update?
Kimi K3.1 Pro has a 1-million-token context window and a Highspeed variant that runs 2-3 times faster. On Artificial Analysis's index it scores 55, fourth overall, while using 150M tokens against a median of 87M.
Moonshot also says it deliberately kept the open weights below its own pipeline's offensive-cyber capability.
Google's wording says it will not use data from these services for ads "unless you choose to make data from these services available to other Google services that support personalized ads."
Commenters argued over what counts as choosing, given how many Google products nag you to turn sharing on.
Disney+ and Hulu Premium went from $18.99 to $21.49 a month on September 23. Mid-show ad breaks stay limited to the cheapest tier for now.
Readers argued about whether any price removes ads for good, or whether the ad-free tier is just the next tier to be sold again.
eMudhra will withdraw from issuing public TLS certificates, keeping revocation services running until every issued certificate has expired or been replaced. Customers get replacements free through partners.
In the thread, a commenter reports that an up-to-date iPhone was still trusting the compromised CAs nearly five days after the report.
Claude Duet is a research preview for US Pro and Max subscribers, powered by Opus 5.5. It runs larger tasks, such as scheduling meetings or building dashboards, on its own cloud computer, and reports back when done.
The first complaint in the thread: a lab with frontier coding agents could let you ask Claude for a spot instead of retyping your details.
Teddy Svoronos says iPadOS 26 behaved correctly: with multiple windows open during AirPlay, the audience used to see only the presenting app. In 27, tapping any other app mirrors everything, poll results included.
Keynote exits the presentation, PowerPoint hangs, GoodNotes recovers. The thread argues over whether hardware deadlines excuse it.
The outage began 19 October at 23:08 PDT; Google called it a power event in a single zone. Per the Neowin headline, HSBC, Revolut, Bank of Scotland, Lloyds and Equifax were hit.
An SRE in the thread says multi-zone setups exist everywhere but zone failure is almost never rehearsed.
Anthropic's report says agents logged six hours of work for every human hour by August, and juniors fell from 27% to 9% of technical hires. Its revenue per employee was 2 to 4 times Meta's and Google's at the same age.
Hacker News mostly distrusted the messenger: the report 'would say that', and one reader pointed out that every line of code is a liability, not a trophy.
Atlas sorts sites into five tiers, from 'activity imminent' to 'no immediate risk'. Sixteen sites in the US, Canada and UK are Tier 1, including EMA4 in Sutton Coldfield and BHX4 near Coventry.
Amazon says it tracks buildings, not individuals, and that Atlas is not used in the UK.
Waymo is partnering with Toronto's Element Fleet Management to run the fleet and says the work will create hundreds of local roles. It cites 400 million fully autonomous kilometres and claims 95% fewer serious-injury crashes than human drivers.
The thread was dominated by riders from other cities saying the same thing: it just works.
Apple's MLX team tested tensor parallelism over RDMA on Thunderbolt 5: four machines decode at 36.73 tokens/s at 8,192 tokens of context, against 15.74 with two-machine pipelining. A single M5 Ultra is about 1.6x faster at decoding than the M3 Ultra.
Meta released Muse Spark Open on October 15 as downloadable weights: 816 billion parameters, 42 billion active per token, a 1M-token context window, under a Muse Model License with limited commercial terms. The benchmark section is empty, and the full evaluation report is "available soon".
Epic Games Launcher V2 preview, which Epic built with AI assistance, starts in about 5.47 seconds against Steam's 5.23 after Valve's quick patch, though Epic still restores from the tray faster (0.36s vs 0.42s).
The stopwatch did not survive HN. A commenter who timed both launchers got 10-15 seconds for a cold Steam boot, and others replied that the faster client had already been sitting in Steam's beta channel.
The support notice says the weekly limit was 50% higher from May 13 through August 19, 2026; it says the promotion is ending rather than introducing a new restriction.
That distinction did little to calm people who had built their coding routine around the temporary headroom and were already reaching limits early in the week.
Apple says alternative distribution still requires notarization, while qualifying web-distributed apps move from a per-install charge to the Core Technology Commission.
Developers focused on the awkward edge case: a customer downloads an app directly, pays for a digital service, and Apple takes a share despite not processing the sale.
Users reported failures across Microsoft services and pointed to GitHub’s long-running move toward Azure; neither observation alone established the root cause.
A second explanation was load: commenters cited rapid growth in commits and new accounts as AI-assisted development turns GitHub into a much busier system.
Google's new Gemini 3.7 Flash is aimed at agent workloads, not just chat. Its launch price is $0.75 per million input tokens and $3.75 per million output tokens.
HN readers did not dispute that it is capable; they asked whether speed and Google's multimodal stack justify the price against cheaper competitors.
The 515-page record makes a familiar loyalty-program bargain visible: small discounts in exchange for a detailed, automatically assembled customer history.
HN’s sharpest correction was uncomfortable, not reassuring. The dossier is unusual to see, but the collection methods are ordinary for large US loyalty programs.
The report examines a Texas data-center project and the power plant planned around it. It makes infrastructure, not model capability, the concrete subject of the story.
HN commenters used the project as a prompt to argue about grid planning, permitting, and whether compute growth is being priced honestly.
Fastmail says it now offers an EU data region. For organizations with residency requirements, the change is useful only if location, migration, support, and operational boundaries are equally clear.
HN readers treated it as a privacy and procurement question rather than a decorative location setting.
The comparison centers on power efficiency, not a single benchmark score. CPU, GPU, display, operating system, and application behavior all change the result a buyer experiences.
HN readers used the article to challenge simple architecture narratives and to ask which workloads still make the trade-off worthwhile.
Taalas was building hardware for fixed AI models, and AMD has acquired it before the company’s first product reached customers.
The thread treated the deal less as a talent acqui-hire than a bet that bandwidth—not compute—is the limiting factor for some inference workloads.
GitHub reported degraded availability for Actions and Pages, interrupting two services many teams use to build, deploy, and publish.
Developers in the thread were less interested in the status page than in the accumulated reliability cost of treating hosted CI as invisible infrastructure.
OpenAI announced ChatGPT updates to GPT-5.6 Sol and broader free-user access to GPT-5.6 Luna.
HN debated whether widening access reflects competition or a more useful product principle: ordinary users should not need to choose a reasoning setting before asking a question.
Google’s leadership announcement combines two large changes: Demis Hassabis is becoming chair of Google DeepMind, while Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le will build Discovery Loop outside Google. Google remains an investor and cloud provider.
HN readers treated the personnel move as the headline, but the sharper question was institutional: how much of Google’s AI research culture is leaving with a group that says it wants to automate scientific and engineering discovery?
A city launch is not the same thing as citywide availability. The operating boundary determines whether a robotaxi is a daily transport option or a useful novelty for a small set of trips.
Readers also framed expansion as a safety and operations problem, not just a map-coloring exercise.
The story is about how hyperscalers finance enormous capital spending, and whether lenders will keep absorbing the paper. That is a financing question before it is a technology verdict.
The discussion’s clearest reversal: heavy borrowing and a debt crisis are not synonyms, especially when the article itself describes investor appetite alongside the warnings.
A retrospective on SwiftUI argues that declarative syntax did not erase the need to drop into UIKit or AppKit. The HN thread added a more practical split: solo developers praise the speed, while complex applications expose the framework’s seams.
A Financial Times report links eBay’s 2019 harassment campaign to a $56 million payout. HN readers returned to the uncomfortable governance question: how did a security operation gain enough internal permission to pursue a couple who criticized the company?
Mike Elgan says Google News now returns social-media pages and results from the wrong country or language, while searches limited to the past week can surface stories that are weeks or months old.
Readers did not merely complain: they named paid alternatives such as Kagi and Ground News, suggesting that basic control over ranking has become a product feature.
The jump is not a single bug-finding demo: Chrome 149 and 150 shipped 1,072 fixes after the team connected model-assisted discovery to triage, patching, testing, and release work.
The uncomfortable constraint is delivery. A public fix can become an attacker’s roadmap before it reaches users, so Google is also piloting two security releases a week and exploring dynamic patching.
A coordinated disclosure describes prompt injection in Copilot for Word: a document can hide instructions in white-on-white text, influence a generated report, then cause Copilot to copy those instructions into the output. That output becomes the next carrier when someone uses it as source material.
The uncomfortable shift is from a bad attachment to a poisoned internal artifact. The author says Microsoft reproduced the broader class after multiple mitigations; the immediate defensive rule is to treat external files as untrusted context, not just untrusted downloads.
Apple Upgrade leases iPhones for 12 or 24 months and now extends the same model to Macs, iPads, and Watches.
The change is not just a new monthly-payment label: keeping the device requires a separate purchase option at the end.
The ruling concerned Google’s attempt to stop scraping through a DMCA theory. Readers saw a narrower legal loss and a broader market signal about access to results after affordable official APIs have become scarce.
The thread’s strongest refrain was not that scraping is consequence-free. It was that restricting the interface while leaving demand intact invites exactly the intermediaries Google is suing.
The proposed ADB change answers a real security concern, but it treats on-device debugging as collateral damage. The author argues that an app cannot silently create the necessary ADB session: the user must already enable debugging and complete pairing or approve a prompt.
HN split between people who see a long-overdue hardening and people who see another capability disappearing from a nominally open OS.
Anthropic released Claude Opus 5 with headline results that put it ahead of Claude Fable 5 on several tests. The release also leans hard on cost per unit of performance, a framing HN immediately noticed.
The thread’s useful split is between benchmark progress and deployment reality: commenters flagged ARC-AGI-3, asked why a new model matters if Fable remains the general-access flagship, and wondered whether safety classifiers alter the practical experience.
The proposal targets on-device ADB, not ordinary USB debugging. Supporters in the thread saw a straightforward physical-security improvement; critics saw another default that removes a power-user option rather than making the risk legible.
The disagreement turns on threat model. An owner with exclusive control of an unlocked phone sees convenience; someone facing an automated inspection or temporary seizure sees an unusually accessible attack surface.
The US government's chief technology officer tweeted that China's Moonshot AI illegally distilled Anthropic's Fable model to build Kimi K3 — released just three weeks after Fable itself went public.
The thread ran the timeline: Fable's access limits lifted July 1, K3 shipped July 16. Commenters found the compressed window less like theft, more like a stretch. One AI-lab-adjacent voice called it "BS to pressure politicians"; HN's own moderator waded in to police the discourse rather than settle the science.
A security researcher: distillation is real, but doesn't require the drama — "99% of the work is already done by that point."
A viral tweet argued passkeys flopped with regular users because the engineers who built them have zero feel for the "consumer brain." The thread agreed passkey UX is broken, then went somewhere the tweet never mentioned.
A security engineer flagged the actual landmine: passkeys support device attestation, letting services silently restrict logins to Google, Apple, or Microsoft's own clients — and there's a maintained list of blacklisted alternative apps.
Someone else pointed out passkeys are conceptually just SSH keys with better marketing — and SSH keys never made it past sysadmins either.
Google's launch post for three new Gemini Flash models compared them only to its own older versions, never once to GLM 5.2, DeepSeek, or any other rival lab.
A Brokk founder ran the missing comparison himself in the thread: Gemini 3.6 Flash lands about even with GLM 5.2 on price and intelligence, just roughly 8 times faster.
Engineers who had built products on the cheap Flash-Lite tier said they were already moving workloads to Deepseek and OpenAI, tired of watching each new release cost more than the last.
A federal lawsuit accused Apple of a design defect for never scanning iCloud for CSAM. In the thread, a commenter who mapped NeuralHash's actual architecture flipped that backstory: the on-device scanner was built specifically to make full iCloud encryption possible, not as an apology after 2021's backlash.
A rival cloud-storage founder in the same thread cut through the debate: client-side or server-side, every scanning method lands on the same "privacy rot," the exact tradeoff Judge Wise says now belongs to Congress, not Apple.
The page that just went live promises ads "clearly labeled" and "separate from answers," rolling out first to only seven countries, about 8% of the world's population, and skipping every paying subscriber.
The asterisk sits in OpenAI's own targeting rules: sensitive chats stay ad-free, but the memory they leave behind doesn't. It can still steer which ad shows up in an unrelated chat the next morning.
Qwen-Image-3.0 launched with demo pages ranging from full newspaper spreads to multi-panel manga chapters and worked exam sheets, claiming a single 4.5k-token prompt and precise Japanese, Korean and Spanish text rendering.
A developer combing through the demo images matched a tell-tale yellow tint straight back to OpenAI's own GPT Image 1, suggesting the model's "authentic details" may have quietly learned from a rival's outputs, a claim nobody at Qwen has addressed.
A Nikkei investigation puts five US tech giants' off-balance-sheet AI debt at $1.65 trillion, an eightfold jump in four years that now tops the $1.35 trillion they report outright, mostly through data-center leases and GPU deals that never touch a debt line.
On HN, the fight moved past the number to who eats the loss if AI demand disappoints: one reply blamed the banks; a sharper reply said the real exposure sits with private credit funds most readers have never heard of.
We Energies won't let Oracle break more ground on its Port Washington campus without $7 billion in collateral: the tariff penalty for any data-center developer whose S&P rating falls below single-A-minus. Oracle's is BBB, two notches under that line, and it just got cut again.
Wisconsin regulators declined to waive the rule, so the guarantee costs Oracle over $100 million a year. In the thread, one member says a friend spent five years on Oracle's own team building its cloud data centers worldwide, then got laid off from it this year.
In a typical single-author contract, half of each payout goes to the publisher, not the writer whose book got pirated — a detail buried inside Judge William Alsup's approval this month.
The creator of books3, the pirated-book dataset that trained a wave of early open AI models, told the thread the real effect: it now takes tens of millions in capital to build a library like Anthropic's, pricing out any hacker with just a laptop.
Two Meta vision models do the labeling: DINOv3 finds the structures, SAM 3 draws the pixel-precise boundaries around them, running across roughly 300 GPUs inside the DOE's Genesis Mission science program.
Meta's post calls the pipeline open source. A commenter who read the actual SAM 3 license found it bans reverse engineering and any weapons-related use, oddly restrictive wording for a federally funded lab.
GPT-5.6 Sol needed 29 logged tool calls to finish; Claude Fable 5 took 54, working the same virtual pencil box, draw, smudge, erase, a peek at the target, inside a benchmark that prices every stroke in tokens and dollars.
The scoring table quietly contradicts the write-up's own verdict: Gemini's similarity score actually beat Sol's on both paintings, yet Sol is the one crowned "runaway leader." A Hacker News reader wasn't buying the piece at all, calling the whole benchmark AI-written.
Buried in Apple's fourth straight quarterly SOC 3 report on the servers behind Private Cloud Compute is a line saying the audit was never built to check whether the AI running on those servers actually works correctly, or keeps a prompt private, only the hardware and provisioning process around it.
An HN member with 22,000+ karma called the badge itself unremarkable: a SOC 3 is just a SOC 2 with the evidence stripped out. Another reader called every audit he's sat through a checkbox exercise.
One reader tested the censorship complaint directly, asking four Chinese models (Kimi K3, two Qwen variants, and GLM 5.2) to write a function named FreeTaiwan(). All four wrote clean, working code with the banned name intact.
A different reader called the whole framing suspicious: the post reads like the argument Palantir's CEO made on CNBC days earlier, and enterprises don't actually care whether a model is open: they care whether the vendor promises to delete their data.
