Hacker News
1Bot traffic from Yandex
Chart reflects only Yandex's own traffic
Many noted the graph shows bot traffic from Yandex's own AS, so a drop is expected when its datacenters go down. Russia-wide and global bot levels barely changed. Yandex runs crawlers and a cloud platform, so most of its datacenter traffic is classified as bots anyway.Yandex Cloud outage and HA lessons
Three strikes knocked out Yandex's failover, disrupting Go, Pay, delivery and Cloud services, including in neighboring countries. Yandex urged customers to switch to competitors. Users discussed military attacks as a disaster-recovery case and how sanctions limit Russian geographic redundancy.Propaganda bots vs. US-dominated bot traffic
Some speculated that propaganda accounts run via datacenter-controlled residential proxies would go quiet, citing fewer pro-Russia comments in Polish media. Others rejected this and pointed out that roughly 40-50% of bot traffic comes from US clouds like AWS, Google and Microsoft.
2The Lightbulb Computer
- Designer Guillaume Ardaud proposes the Lightbulb Computer, a projector with computer vision in a lightbulb shape that projects interactive information onto real-world surfaces, offering a more social, less distracting alternative to screens and face-worn headsets.
Prior art and similar concepts
Commenters cite earlier attempts: Samsung's projector phone, Amazon's patent, IKEA/IDEO's future kitchen, Dynamicland/Folk Computer, Microsoft/Lenovo table computers and Johnny Lee's Wii work. Others argue revisiting old ideas with better tech like LLMs is worthwhile.Privacy and data control concerns
Some find an always-watching, always-listening home "jinn" unsettling, especially in private rooms. Many would only accept it if fully local and open-source, fearing commercial versions would be cloud-based surveillance like Flock.Technical feasibility and form factor
The author explains the Mac, 4K laser projector and webcam setup. Debate covers ambient light, projector efficiency and the bulb form factor, with one arguing smart glasses are better. Others suggest latency and gorilla-arm fixes or describe DIY builds.
33rd Yandex Cloud Data Center Was Hit
Strikes as retaliation
Commenters say Russia hit Ukrainian telecom and data centers first, making this payback. Others note Russia's digital sovereignty push leaves it with vulnerable domestic targets, while Ukraine can rely on Western clouds. This complicates Europe's debate on leaving US providers.Protecting data centers from attack
Users speculate data centers will move to bunkers, mines or Siberia, but cite cost, heat rejection and exposed power/cooling links. Orbital data centers are dismissed over radiator physics and debris or anti-satellite risks. Some argue data centers are low-priority targets hit with spare drones.Responsibility of ordinary Russians
Some hope the strikes bring the war home. Others say ordinary Russians oppose it but risk 15-year sentences. Critics counter that many support or justify the war, serve in the army, or tolerated Putin's decades-long rise.
4A city-building game in which the city would prefer you didn't
- Discretionary Review is a simulator about housing in San Francisco.
Gameplay strategies and reactions
Players share scores and build orders: market-rate projects in friendly districts, the Mission Rock megaproject, then affordable housing around 2020. They praise the dark humor and real proposal sites, while noting investor return demands make winning nearly impossible.SF density, zoning and transit housing
Commenters ask whether SB 79 helps and propose towers atop Mission BART stations, which have been blocked for a decade. They debate whether 10 floors is "huge" versus Paris and London, and whether SF effectively wants a hukou-style residency system.HS2 and infrastructure costs
The UK's £100M HS2 bat tunnel is debated. Some say wildlife law required some fix and the tunnel was the cheapest option, while others cite 8,500 consents. Many blame costly tunnels on placating Conservative voters and prestige-driven very high speed requirements.
5Build your own decision model
- "System one" decision models choose from fixed answer options in a single pass instead of generating token by token. A small Qwen model shows this works and improves with fine-tuning, but its confidence scores are overconfident until calibrated with temperature scaling.
Jev hype vs. traditional classifiers
An MLE vented that classifiers were ignored for years until the Jev hype. Others said its if/switch-style API, low cost, speed and zero-shot convenience make it genuinely new. Skeptics called decision models rebranded classifiers, or fine-tuned ModernBERT dressed up in mysticism.Real-world uses of decision models
Users shared uses like sorting email into folders, advanced autocorrect and filtering text chunks to cut LLM token use 10x. Others flag dangerous commands in Claude Code hooks, build game test harnesses or classify research papers. Speed and low cost were the main draws.Calibration and evaluation concerns
Commenters likened tuning temperature on a benchmark to appear calibrated to p-hacking and test-set overfitting. Others warned against assuming AI decisions are accurate without evaluation, noting small trained classifiers often beat Jev on fixed tasks like banking77.
6Nitter: Update Oct 10th seeking funding and legal help
- X Corp. sent cease and desist letters in August and September 2026 demanding Nitter's permanent takedown. An earlier claim of legal representation proved untrue, and Nitter has no counsel. The developer seeks funding and legal help, with donations going first to legal costs.
Whether nitter can legally operate
Some compare Nitter to proxying NYT and doubt it can run legally. Others say X doesn't own user content and ToS aren't binding without agreement. Counterpoints: ToS is contract law, CFAA poses risks, and X could gain copyright standing via exclusive content deals.Legal intimidation and ways to resist
Many see X's action as harassment meant to chill speech against an underfunded developer. The creator says a supposed pro bono lawyer misled him. Others suggest contacting the EFF, ignoring C&Ds, or Pirate Bay-style distributed proxies, though losing the domain could kill the project.Purpose of nitter and X's relevance
Nitter lets people read public posts without an account, ads or tracking, using logged-in server sessions. Commenters call X a gated "town square" still used by journalists and officials. Some say just don't use X, while others want its relevance ended entirely.
7LineageOS 24
- LineageOS 24, based on Android 17, arrives early ahead of the project's 10th anniversary in December 2026. It brings security patches, refreshed apps including a new image editor, browser-based flashing tools and an accessibility audit, while older legacy devices remain on version 22.2.
Keeping old phones alive
Users praise LineageOS for extending the life of the Pixel 4a, OnePlus 6T/7 and Poco F3. Others warn that EOL devices keep unpatched modem/baseband firmware, blobs and kernel CVEs. Battery costs, banking apps and camera quality also push upgrades.LineageOS versus GrapheneOS
Some prefer GrapheneOS for storage/contact scopes, hardening, a relockable bootloader and fast Android 17 updates. Defenders note LOS is volunteer-run, supports far more devices and gives users more control. Consensus: GOS on recent Pixels, LOS elsewhere.Generic bare-metal Android target
An experimental target booting on x86 PCs, Apple Silicon Macs, DGX Spark and Snapdragon laptops excites users. Skeptics ask why use Android over Linux. Replies cite better sandboxing, familiarity, phone-only apps, games and x86 tablets.
8Bitwarden Dual License Model
- Bitwarden will publish commercially licensed builds in app stores and on its download page. The GPLv3 version stays on GitHub, and current features remain in both. Some future features may be commercial-only. Users voiced concerns about a gradual move away from open source and about self-hosting.
Private equity and price hikes erode trust
A blog post on Bitwarden's PE-linked new leadership convinced many to leave. Users cite the price nearly doubling ($10 to $19.80/yr) and fear "boil the frog" tactics. Some dismissed the post as LLM-written, and others argued free users shouldn't expect much.Searching for alternatives
Suggestions include self-hosted Vaultwarden with the Keyguard client, KeePassXC synced via cloud storage, Proton Pass, Enpass and AliasVault. Many also called Bitwarden's clients slow and buggy, while others warned self-hosting carries its own security burden.License change and open source sustainability
Some accept source-available licensing to stop big clouds leeching, citing Elastic and Redis. A former Redis employee described failing to push leadership toward a gentler relicense. Others call relicensing greed and fear it's a first step toward closing off features.
9Documents add to evidence in climate deception case against ExxonMobil
- Newly released documents from Massachusetts' lawsuit against ExxonMobil show the company foresaw climate action threatening its business in 1988. Its own scientists called its algae biofuel and carbon capture claims false or overstated, while Exxon denies misleading the public.
Accountability for executives
Many doubt anyone will be jailed, citing wealth, diffuse responsibility and political capture. Some propose a corporate death penalty; others argue executives and boards, not employees, should bear personal liability, invoking "piercing the corporate veil."Is the memo sensationalized?
One commenter says the 1988 memo is far milder than the headline and defends oil-company carbon capture researchers as acting in good faith. Critics call the memo damning, the defense a strawman, and carbon capture a tool for enhanced oil recovery.Shared blame vs. deliberate deception
Some argue the risks were public knowledge from academic research and that consumers and politicians share blame. Others reply that Exxon actively manufactured doubt like Big Tobacco, citing "Merchants of Doubt" and the reported accuracy of Exxon's own predictions.
10I paid people to try and follow my README
- To beat his own biases, a developer paid testers €25 each to install ActivityBot while thinking aloud, uncovering wrong links, confusing jokes, jargon and missing explanations. Revising after each session improved the README, showing developers should test docs with real people.
Curse of knowledge in documentation
People find their own notes become incomprehensible within weeks once context fades. Advice: write for a tired, older future self, use "how to make a sandwich" detail, list prerequisites, and improve docs through dogfooding or even LLM checks.READMEs that never explain what things do
Many complain that READMEs and landing pages use buzzwords or "X is an alternative to Y" without saying what the product does. Some say vague marketing targets finance approvers. Others urge making docs the homepage, adding screenshots and defining abbreviations.Parallels with UX usability testing
Commenters liken paying people to follow a README to UX usability testing, citing Nielsen Norman Group research that about 5 users uncover most issues. Experts often write for other experts. AI agents can help test docs but fail confidently instead of asking "wait, what?"