Hacker News, Distilled

Recyclability Claims & Greenwashing

• Many see Starbucks’ “widely recyclable” claim for #5 polypropylene cups as classic greenwashing, given that most US municipalities don’t actually recycle #5.
• Label is tied to an industry-backed “How2Recycle” system with little apparent regulatory oversight.
• Several commenters stress the gap between “accepted in the bin” and “actually recycled,” calling the public messaging misleading.

Tracker Study & Methodology Debates

• Some argue the Bluetooth-tracker study is more of an advocacy stunt than a rigorous audit:
  • Trackers themselves are non-recyclable and may be intentionally removed or diverted.
  • Many trackers stopped pinging on highways; a few went to known recycling-related facilities but were excluded from the headline claims.
  • Transfer stations can also handle recycling, so “went to transfer → landfill” is seen as an unsupported leap.
• Others counter that, even if imperfect, the study highlights how consumer-facing “recyclable” claims don’t match the likely end fate of items.

Economics & Practical Reality of Recycling

• Multiple accounts describe systems where everything is collected as “recycling,” then sorted; only materials with buyers (notably aluminum) are actually recycled, the rest landfilled or exported.
• Plastics (especially beyond PET/HDPE) often have negative economic value as feedstock.
• Glass and paper recycling are technically feasible but heavily dependent on local infrastructure, contamination, and transport costs; glass is often downcycled.

Plastics vs Other Materials

• Strong consensus that plastics are the hardest to recycle economically and environmentally; some argue we should landfill all plastics and use only virgin plastic when needed.
• Aluminum is praised for high recyclability but criticized for energy intensity; recycling aluminum is still said to be far cheaper than new production.
• Glass is seen as highly recyclable in principle, but only when pre-sorted and locally processed.

Individual vs Systemic Responses

• Several commenters say recycling has become a low-impact “feel-good” activity that distracts from bigger levers like reducing driving, energy use, and consumption.
• Others emphasize regulations, extended producer responsibility, and bans/mandates (e.g., on single-use packaging) as the only realistic way to change corporate behavior.

Reuse, Composting, and Landfills

• Starbucks already allows (and sometimes incentivizes) bring-your-own cups, but uptake is low due to inconvenience and habit.
• Some municipalities offer composting for soiled paper (e.g., pizza boxes); many do not.
• A minority argue modern landfills are acceptable and that, in some sense, landfilled plastics act as carbon sequestration.

Positioning vs. Flipper Zero

• Widely seen as a different product, not a direct successor.
• Flipper Zero was a “toy-like” RF gadget with NFC/RFID/IR/sub‑GHz built in.
• Flipper One is viewed as a compact Linux cyberdeck / portable computer with networking focus.
• Some lament loss of built‑in RF features; others argue Zero’s RF was limited compared to dedicated tools.

Hardware & Design Choices

• High-end SoC (A72/A53) with 8 GB LPDDR5 leads people to compare it to Raspberry Pi, NUCs, Steam Decks, tiny laptops.
• Dual gigabit Ethernet, Wi‑Fi, HDMI/DisplayPort, M.2 slot and SIM slot draw praise, especially for network work.
• The small monochrome transflective-style display is controversial: criticized as “crappy” for the power level, but defended as sun‑readable and low power.
• Screen is driven by the microcontroller, with Linux seeing a framebuffer; this enables MCU-side overlays, low‑power modes, and recovery if Linux hangs.

Radios, Expansion & Regulation

• Many are disappointed by the lack of built‑in NFC, RFID, sub‑GHz RF, IR, and 1‑Wire.
• M.2 expansion (including SDR modules) is seen as powerful but expensive and making RF effectively “required accessory” rather than core.
• Some suggest RF was externalized to avoid regulatory/customs issues and keep the base device safer to sell globally.

Price, Alternatives & Usefulness

• Speculation ranges from ~$250–400 to $500–1000+.
• Several argue that above ~$400 it becomes hard to justify against cheap laptops, Steam Decks, GPD devices, or a HackRF plus computer.
• Others think a premium “Swiss army knife for networks/cyberdeck” can command a higher price for enthusiasts.

Potential Use Cases

• Frequently mentioned: travel router, mobile router (Ethernet + Wi‑Fi + cellular), inline MITM/sniffer, VLAN/DHCP/PXE diagnostics, WoL helper.
• Also discussed: attach SDR via M.2, run local LLMs/agents using the NPU and IO, “shit‑hit‑the‑fan” cyberdeck, AI/voice‑controlled scripting via PTT.

Concerns & Skepticism

• Doubts about battery life and heat relative to the Zero’s weeks-long standby.
• Some think the spec sheet and marketing copy look AI-generated or unfinished (“needs verification/clarification”), reducing trust.
• Rockchip’s binary blobs and GPL issues are raised; claims of “better mainline Linux support” are met with skepticism.
• Size, weight, appearance, and possible scrutiny/confiscation by border/TSA agents are minor but recurring worries.

Role and legacy of Qian Xuesen

• Seen as central to China’s missile and aerospace programs and to long-term organizational capacity-building, though some argue his work in China was largely managerial by then.
• Also credited with co-founding major institutions (e.g., JPL, a top Chinese science university) and being a strict, high-standard educator.
• Debate over how much he uniquely accelerated China’s rise versus being one capable node in a capable system.

Was deporting Qian a US strategic blunder?

• One camp calls his imprisonment/deportation a major strategic mistake and symptom of abandoning empiricism and pragmatism during the McCarthy era.
• Others argue decisions must be judged given uncertainty: China likely would have developed similar capabilities within roughly a decade anyway, so at worst the US accelerated Chinese missiles/aviation by some years.
• Some emphasize the real counterfactual is not the prisoner swap but not imprisoning him in the first place.

Communism, McCarthyism, and security concerns

• Thread notes Qian’s attendance at Communist Party–linked meetings, refusal to testify against a colleague, and early security concerns predating McCarthy.
• Dispute over whether he was a committed communist, a pragmatist caught between powers, or simply a nationalist who came to believe in Mao.
• Strong disagreement on framing: some stress pervasive Red Scare overreach; others highlight extensive real Soviet infiltration to argue anti-communist fears weren’t purely paranoid.

Immigration, xenophobia, and talent flows today

• Qian’s story is seen as a cautionary tale: how many high-talent people are now leaving or never coming to the US due to xenophobia, complex visas, or anti-Chinese sentiment.
• Anecdotes of Chinese and other foreign STEM graduates pushed out by visa hurdles and then contributing elsewhere (e.g., China, Canada).
• Discussion of both right-wing and progressive forms of anti-Asian bias; some point to espionage cases and PRC diaspora influence as security concerns.

Qian’s broader influence and misjudgments

• Praised for early advocacy of new energy vehicles and AI.
• Criticized for pseudoscientific “superpower” promotion and for an overoptimistic agricultural yield estimate that may have influenced Great Leap Forward policies; responsibility and political context are contested.

Representation in media and narratives

• Debate over why Americans haven’t made a major film about him, unlike Oppenheimer or tech founders; some note there are multiple Chinese films/series.
• Wider discussion on how history overemphasizes lone geniuses versus institutions, funding, and teams.

Reactions to the article’s anti-LLM stance

• Many readers think the critique overgeneralizes from limited experience with weak or free models.
• Others resonate with the discomfort at “paying to think” and the desire to avoid SaaS lock‑in.
• Some appreciate the writing and the focus on process over product, even while disagreeing with the conclusion.

Productivity vs. craft and “hard problems”

• One camp argues LLMs automate “lower-tier” mechanical coding, freeing humans for higher‑level design and more complex systems.
• Another camp feels core, enjoyable parts of engineering are being offloaded, weakening skills and understanding.
• Disagreement over whether recent typical dev work was truly “hard” or mostly framework/config glue.

Spectrum of LLM use (beyond vibecoding)

• Several commenters reject the binary of “no LLMs” vs “agent writes everything.”
• Common “middle ground” uses: autocomplete, one-off snippets, boilerplate, tests, integration glue, while humans review every line.
• Others report using agentic tools heavily but still steering architecture and reviewing output.

Costs, access, and “cheapskate” ethos

• Strong current of people who avoid recurring SaaS fees and prefer FOSS and local tools; LLM subscriptions feel culturally wrong, not just expensive.
• Counterpoint: $20–$100/month is seen as trivial relative to productivity gains, especially for startups.
• Concern that rising and opaque token costs could make experimentation and hobby work less viable.

Code quality, maintainability, and complexity

• Some see LLMs enabling faster delivery of working systems and personal projects that would otherwise be infeasible.
• Others report AI‑written codebases as sprawling, incoherent, and harder to reason about than hand‑written code.
• Fear of becoming dependent on tools to maintain code they generated; worry about “deskilling” and bloated, low‑quality output.

Agentic environments and local models

• Enthusiasts emphasize that results depend heavily on the “harness”: sandboxing, tooling, context strategies, and multi‑agent workflows.
• Local/open‑weight models are seen as a path to reduce cost and lock‑in, though performance and hardware demands are debated.

Analogy and culture wars

• Recurrent analogies compare LLM refusal to refusing cars or tractors; critics call this a “luxury belief,” supporters note external costs.
• Some frame coding-without-LLMs as “trad coding” or a kind of identity/virtue choice, for better or worse.

Media coverage & sensationalism

• Debate over whether the article is sensationalistic or useful.
• Some argue it cherry-picks a small-town outburst to confirm reader biases and overplays a “crash out” narrative.
• Others say amplifying local reporting strengthens accountability norms and can inspire similar policy moves elsewhere.
• There’s a broader concern that consistently selecting the most dramatic Flock-related stories can distort public understanding, even if factually accurate.

Value of small-town surveillance debates

• Disagreement over whether events in a town of ~800 people are worth national attention.
• Critics see limited policy relevance; supporters say local votes on Flock are concrete examples for other councils to learn from.

Surveillance, safety & deterrence

• Some say surveillance doesn’t prevent crime, only aids prosecution after the fact.
• Others counter that deterrence hinges on perceived likelihood of apprehension; cameras and ALPRs arguably increase that, even if prison length matters less.
• Counterpoint: more cameras don’t automatically translate into more arrests or prosecutions; enforcement capacity and prosecutorial will still matter.
• There’s concern that marginal deterrence gains may not justify large privacy trade-offs.

Privacy, civil liberties & hypocrisy

• Strong pushback against “if you have nothing to hide, you shouldn’t care” arguments.
• Commenters note this presumes a just system and ignores risks from bad judgment or malicious use.
• Several point out the hypocrisy: many who support broad surveillance would likely object to a camera pointed at their own home.

Local politics, lobbying & bribery

• Dispute over whether it’s plausible that small-town councilmembers are bribed or influenced by vendors like Flock.
• Some say big companies wouldn’t bother with a tiny market; others cite examples of small-town corruption and argue that influence can be very cheap (donations, perks, social attention).
• A middle view notes a spectrum: from illegal bribery to legal-but-shady lobbying to simple relationship-building.

Councilmember’s “modest proposals” response

• The satirical proposals to ban phones, outward-facing cameras, and even internet/records are widely interpreted as a tantrum after losing the Flock vote.
• Some see it as a standard, if performative, rhetorical move (reductio ad absurdum); others call it false equivalence and bad faith.
• Many believe it reveals an all-or-nothing mindset and poor representation of constituents’ clearly stated opposition to Flock.

Scope of the Incident & Railway’s Profile

• Railway reported that Google Cloud automatically suspended its production account, making persistent disks inaccessible and later restored, implying suspension rather than data loss.
• Railway claims Google told them it was an incorrect automated action affecting many accounts.
• Debate on how “high-profile” Railway is: some see them as well-known in dev circles with significant usage stats; others see them as a small, noisy startup.

Should Google Issue a Public Statement?

• Many argue Google should explain what happened for PR, trust, and risk-management reasons, especially given the platform-wide nature and impact on Railway’s customers.
• Others say Google likely cannot or should not disclose customer-specific details without consent, citing B2B confidentiality norms.
• Some suggest the right path is: Google explains to Railway; Railway decides what to share—or publicly states if Google refuses.
• There is concern that mandatory arbitration and NDAs will keep details opaque; some argue courts and public records would be healthier.

Automated Suspensions, Support, and Platform Risk

• Strong criticism of Google’s heavy reliance on automated enforcement with few human escalation paths, even for sizable paying customers.
• Multiple anecdotes of abrupt GCP suspensions (e.g., missed verification email) causing extended outages and slow remediation.
• Users worry that if a customer as large as Railway can be taken down without warning and no immediate human contact, smaller startups are even more vulnerable.
• Calls for Google to:
  • Publish how suspension decisions are made.
  • Exempt large/critical business accounts from fully automated shutdowns, requiring human review and proactive outreach.

Who Is at Fault?

• Some suspect Railway’s PaaS model (hosting spammers/malware, weak abuse controls, shared infra) may have triggered abuse systems legitimately.
• Others argue that rapid reinstatement and Railway’s account of an “incorrect” action points to Google error.
• General agreement that Railway’s architecture—allowing a single provider action to cascade into a platform-wide outage—was a serious design flaw, which they themselves acknowledge.

Broader Cloud Provider Trust Comparisons

• Many express loss of confidence in GCP specifically, despite praising its technical quality and security.
• AWS is cited as more trustworthy due to responsive, proactive support.
• Azure receives strong negative sentiment from several commenters.
• Some prefer multi-cloud, bare metal, or treating any single-cloud deployment as disposable due to platform risk.

Context: Pokémon Central Wiki Deindexing

• Italian Pokémon wiki reports going from ~500k indexed pages to 11, with >100k URLs marked “crawled but not indexed.”
• Other small wikis and long‑running blogs report similar recent drops and “crawled but not indexed” with no clear reason in Search Console.

Speculated Technical Causes

• Possible bugs or “jank” on Google’s side; large systems can accidentally exclude small fractions of sites.
• Hypotheses: Cloudflare anti‑bot rules blocking Googlebot in some paths; wiki spam or malware; misconfigured robots.txt (though OP says this was checked); use of anti‑LLM training flags (e.g., TDMRep) coinciding in time.
• One theory: because much content is translated from an English wiki, internal systems may treat it as easily derivable and deprioritize crawling/indexing.
• Another angle: brand confusion with “Pokémon Trainer Central” rebrand affecting rankings for “Pokemon Central,” though OP stresses the bigger issue is loss of indexing for specific topic pages in Italian.

Wikis, Spam, and Anti‑Abuse

• Multiple wiki maintainers describe severe modern spam: bots, “sleeper” accounts, LLM‑assisted sign‑ups solving CAPTCHAs.
• Mitigations mentioned: Cloudflare rules, limited permissions for new accounts, manual patrolling, Anubis (anti‑scraper), invite‑only systems, custom knowledge‑based CAPTCHAs, domain blacklists.
• Debate over tree‑based invite/reputation systems: some see them as powerful; others point to abuse, account hacking, and raising barriers for genuine newcomers.

Broader Indexing Trends

• Several participants note widespread “crawled but not indexed” reports and argue Google is drastically shrinking its index to a smaller set of “primary authorities.”
• Some think wikis are inherently high‑risk SEO targets and require meticulous hygiene (sitemaps, metadata, spam control).

Economic and Power Concerns

• Sites relying on Google Search + AdSense see sharp traffic and revenue hits; some move to in‑house ads, acknowledging significant extra work.
• Strong criticism of opaque, one‑sided decisions: platforms can effectively “disappear” sites or accounts without clear explanation, functioning as unaccountable gatekeepers.

Alternatives and Changing Search Behavior

• Kagi, DuckDuckGo, Brave, Startpage, Ecosia, Yandex, Marginalia cited as alternatives; some report better results, others mixed.
• Several argue classic SEO and search traffic are fading as users increasingly ask LLMs directly; others note LLMs still depend on SEO‑shaped web data.

Views on Google’s Motives

• Split between “bug/latent side effect” vs. “deliberate strategy” interpretations.
• Many describe Google as profit‑maximizing, indifferent to publishers, pushing zero‑click results, AI overviews, and more ads.
• Some call for antitrust or EU action; others see this as the predictable evolution of an ad‑driven monopoly.

Overall Market & Bubble Context

• Many see the IPO as a late-stage move in an AI bubble, likening it to the dotcom era and Netscape’s IPO as a potential trigger for a final run-up before a crash.
• Others argue we may already be closer to the peak: high Nasdaq P/E, banks offloading discounted data‑center loans, VC liquidity constraints, and general macro anxiety.
• Some think OpenAI/Anthropic/SpaceX “trillion‑dollar IPO summer” could stretch markets further; others predict one of these IPOs will flop and mark the start of a downturn.

OpenAI Financials & Business Model

• Reported revenue figures (tens of billions annualized, up sharply year-over-year) are debated against huge capex and training costs.
• Some claim each new model brings in revenue multiples of its cost; skeptics note scaling laws, rising marginal costs, and thin margins at peers.
• A recurring theme: “If the unit economics were truly that good, they’d raise debt, not equity.”
• The CFO has reportedly said internal systems aren’t ready for full public reporting until 2027, fueling doubts about the quality of forthcoming disclosures.

IPO Mechanics, Liquidity & Index Funds

• Strong view that late IPOs primarily provide exit liquidity for early insiders; others counter that history shows substantial post‑IPO upside can still exist.
• Concern that shortened index-inclusion timelines mean S&P/Nasdaq trackers and pension funds will be forced buyers at peak valuations, potentially becoming “bag holders.”
• Debate over how much retail vs institutions actually drive IPO pops and who ultimately bears losses.

Competition, Moats & Open Models

• Several argue OpenAI is no longer the clear product leader; Claude and Gemini are often cited as superior on capability or tooling, though OpenAI still wins on brand and ease of API use.
• Open‑weight models (e.g., DeepSeek) are seen as rapidly closing the gap at far lower cost, pushing commoditization and questioning any lasting moat.
• Others respond that infra, scale, CUDA-like ecosystems, and enterprise integration are still meaningful barriers.

Ethics, Governance & Nonprofit Origins

• Strong criticism of the shift from original nonprofit, “for the public good” mission to a highly financialized, closed, for‑profit structure.
• Some fear public ownership will further prioritize short‑term returns over safety, R&D, and openness.
• A minority express optimism or indifference, focusing on profit potential rather than governance or societal impact.

Settlement Size and Adequacy

• Many think ~$835k is low for 37 days in jail, especially given the uncertainty and fear of “indefinite” detention.
• Others say it’s very high relative to typical personal-injury payouts and unproven economic damages.
• Some note attorney fees and taxes may significantly reduce the take‑home amount; others point out FIRE’s work is pro bono.

Non-Monetary Harm

• Commenters stress the key harm was not “37 known days” but not knowing when or if release would come.
• Additional harms cited: job loss, missed life events, ongoing harassment from political opponents.

Who Pays: Taxpayers vs Officers

• Strong frustration that local taxpayers, not the sheriff or investigators, will likely fund the settlement.
• Proposals:
  • Make officers personally liable, or require individual malpractice-style insurance.
  • Charge settlements to police pension funds or department budgets to align incentives.
• Counterpoint: direct government liability is appropriate because the abuse flowed from official authority and can pressure systemic reform.

Qualified Immunity and Accountability

• Qualified immunity is widely criticized as blocking meaningful civil accountability for officials.
• Some argue this case is a clear, “knowingly” unconstitutional act where immunity should not apply.
• Others note that legally, false imprisonment/kidnapping generally don’t attach once a warrant and “due process” exist, even if later found unconstitutional.

Systemic Reform vs Criminal Punishment

• One camp: officers (and possibly judges) should face criminal charges for such rights violations; otherwise abuse will continue.
• Another camp: the U.S. already over‑incarcerates; better to reduce prosecutorial/police power (tighter warrant standards, automatic prompt bail hearings, end cash bail abuse, expand civil remedies) rather than add new criminal statutes that could be weaponized.

Free Speech, Memes, and Comparisons

• Broad agreement that reposting an accurate meme critical of a politician is core protected speech.
• Some compare to cases where misleading election memes were prosecuted; others distinguish them as intentional fraud versus accurate political commentary.
• Thread contrasts the U.S. First Amendment environment with UK/EU “harmful” or “grossly offensive” speech laws, noting more arrests there for online posts.

Role of Sheriffs, Judges, and Local Politics

• Discussion emphasizes the sheriff’s elected status and the magistrate/judge’s role in approving an obviously unsound warrant and excessive bail.
• Concern that small-town “fiefdoms” and weak local media oversight let similar abuses happen without national attention.

Attack vector and scope

• Breach tied to a compromised VS Code extension, identified in the thread as “nx console,” later confirmed via GitHub’s own blog and the extension’s security advisory.
• Malware on an employee’s device led to unauthorized access to ~3,800 internal GitHub repositories; commenters note this is likely a subset of total internal repos.

How exfiltration likely worked

• Consensus: the extension/malware harvested local secrets (SSH keys, PATs, env vars) and exfiltrated a small encrypted payload to attacker-controlled infrastructure.
• With valid tokens, attackers could clone any internal/private repos those credentials could access, without re-auth or 2FA, and often without triggering effective alarms.
• Commenters stress that preventing exfiltration from an internet-connected dev machine is “virtually impossible,” and detection is hard if access looks normal.

VS Code and extension security concerns

• Heavy criticism that VS Code has “no real security model”: extensions, front-end, and back-end share broad, unsandboxed access.
• Several people highlight long-standing, unresolved requests for an extension permission system and sandboxing.
• Others argue this is a general problem for any extensible editor or plugin ecosystem, not VS Code-specific.

Mitigations discussed

• Network-level: restrict outbound connections (per-app firewalls like OpenSnitch, allowlists), monitor unusual traffic, especially to nonstandard domains.
• Platform-level: sandbox IDEs/extensions (WASM/WASI, containers, Flatpak-like models), limit file-system visibility, block or gate network access.
• GitHub org controls: enforce SSO, IP allowlists, PAT restrictions/expiry, audit log streaming, and collection of HTTP logs; use canary tokens and static analysis for Actions.
• Personal practices: disable auto-updating extensions, minimize extension count, prefer “official” or self-written ones, and use delayed adoption of new package versions.

Ransom and leak dynamics

• Attackers reportedly offered the internal repos for a minimum of $50k.
• Debate over whether paying ransoms ever makes sense: some argue experienced groups will honor deletion to preserve their “business model”; others insist there is no credible way to verify deletion and paying only adds risk and cost.

Broader reactions

• Some call for moving away from VS Code and GitHub, or from Microsoft ecosystems generally.
• Others note that large orgs will inevitably accumulate thousands of repos and that extension/package supply-chain attacks will become more common.

Required exercise time and benefits

• New guideline (560–610 minutes/week of moderate–vigorous activity) is contrasted with prior ~150 minutes/week.
• Study claims ~8–9% cardiovascular risk reduction at 150 minutes vs >30% at ~10 hours/week.
• Some note other umbrella reviews suggesting much lower volumes (e.g., 15 MET-hours) already capture most benefits, making this result seem extreme.
• Several emphasize that smaller amounts still help; “substantial” is a definitional choice.

Feasibility and life constraints

• Many find 9–10 hours/week unrealistic, especially for parents in dual‑income households.
• Others argue it’s possible by:
  • Integrating exercise with childcare (stroller runs, playing, hikes).
  • Active commuting (cycling/walking to work).
  • Small, consistent habits (e.g., 30 seconds of daily calisthenics with kids, then expanding).
• Tension between “people make excuses” and recognition that time, job, kids, climate, and housing strongly constrain options.

What counts as “moderate” or “vigorous”

• Confusion over definitions: brisk walking is classified as “moderate,” vigorous is framed as sustained higher heart‑rate zones, distinct from all‑out HIIT.
• Debate on whether chores, normal walking, and weight lifting count; some wearables show everyday walking barely raises heart rate.
• Clarifications from the thread:
  • Walking, housework, gardening often counted as moderate.
  • Vigorous minutes may be weighted more (e.g., 2×) in guidelines.
  • HIIT is described as beyond “vigorous” and not sustainable at high weekly volumes.

Health tradeoffs and human limits

• Some argue 10 hours/week is too big a time cost relative to added lifespan; others highlight healthspan and enjoyment (e.g., sports people love).
• Concerns about joint wear and injuries; countered by claims that the body evolved for regular movement and most people under‑exercise.
• Skepticism about very high volumes, especially in older ages and with potential overtraining.

Study design, bias, and uncertainty

• Study is observational; several criticize causal language and note correlation vs causation issues.
• Participant profile: average age ~57, mostly white. Some see this as “late in life” and not broadly representative.
• Potential confounders: people who exercise more may also eat better, avoid smoking, have more time and money, and care more about health.
• Use of accelerometer data:
  • May miss activities with little wrist movement (cycling, some strength work).
  • Raises questions about how “moderate/vigorous” minutes were inferred.
• Exclusion of very high VO2max values as “implausible” and undercounting of certain activities are flagged as possible flaws.
• Overall sentiment: mixed—some find the results motivating or validating; others view the headline as overreaching or discouraging.

What Wero Is

• Pan‑European payment initiative (EPI) built on top of SEPA Instant (SCT Inst).
• Primarily a UX and alias layer: maps phone numbers (and similar IDs) to IBANs and triggers instant SEPA transfers.
• Consolidates or replaces existing national schemes: iDEAL (NL), Paylib (FR), Bizum (ES), BancomatPay (IT), SIBS, Vipps/MobilePay, etc.
• Roadmap: P2P now, wider e‑commerce and PoS/merchant support targeted around 2027.

How It Works & Current Adoption

• Integrated mostly into existing banking apps; sometimes a dedicated Wero app.
• P2P: send money using phone numbers; recipient often doesn’t need prior registration if their bank participates.
• Online: merchant shows Wero/iDEAL/Bizum option → user selects bank → redirected or QR scanned → confirms in bank app.
• In‑person: QR codes today; some early contactless support via national systems (e.g., Bizum terminals, Swish/Vipps‑style flows).
• Reported heavy real‑world use in France (ex‑Paylib), the Netherlands (ex‑iDEAL), Spain (Bizum), and other local schemes; others say their banks still don’t support it.

Benefits & Positive Experiences

• Instant and usually free P2P across banks and, eventually, borders.
• No card numbers on merchant sites; bank handles authentication (often via app + biometrics).
• Less friction in splitting bills and small payments; users like “just use my phone number.”
• For online merchants, can be easier and cheaper than card acceptance once integrated.

Limitations and Critiques

• Functionally close to “SEPA Instant + phone aliases”; some see it as underwhelming vs PayPal (buyer protection, dispute handling, IBAN obfuscation).
• Chargeback / dispute layer is weaker or unclear compared to card schemes.
• Adoption uneven: some major banks and regions lag; bank apps often clunky.
• Smartphone‑only orientation, QR codes, and occasional contact‑sync requirements raise usability and privacy concerns.
• Not a full card network: no pre‑auth, card‑on‑file semantics, or credit features yet.

Impact on Visa/Mastercard & Merchants

• Many consider “Goodbye Visa/Mastercard” overstated: cards remain dominant for in‑store contactless and international travel.
• Real near‑term impact is on domestic online and P2P flows; card rails still back many debit cards.
• Merchants may gradually prefer cheaper Wero‑based payments, but replacing entrenched POS infrastructure is seen as hard.

Sovereignty, Politics, and Infrastructure

• Strong framing as European “payment sovereignty” and diversification away from US‑controlled rails amid tariffs, sanctions, and political volatility.
• Debate over whether central‑bank or bank‑run systems are preferable to US card duopoly; also fears of future CBDC‑style overreach and surveillance.
• Some note irony that parts of Wero run on AWS and depend on Apple/Google platforms, questioning how “sovereign” it really is.

Scope of Meta’s Actions

• Meta is reportedly blocking human-rights–related accounts from audiences in Saudi Arabia and the UAE; the site of one NGO is itself blocked in the UAE.
• Some note Meta has similarly removed or limited rights-related accounts in democratic countries (e.g., LGBTQ groups in the Netherlands).

Obeying Local Law vs Moral Responsibility

• One view: Meta “has no choice” but to follow local laws where it operates; otherwise it risks shutdown, blocked traffic, or staff persecution.
• Counterview: There are clear alternatives:
  • Exit those markets entirely.
  • Refuse and let regimes build their own firewalls.
• Critics argue compliance makes Meta complicit in human-rights abuses, not merely “amoral,” and that profit and shareholder pressure drive this.
• Others stress that the underlying problem is repressive governments; Meta is just responding to incentives.

Corporate Power, Politics, and Double Standards

• Debate over whether US/EU governments are themselves deeply complicit (weapons sales, surveillance, alliances with Gulf states) and thus not credible moral arbiters.
• Repeated claim that large tech firms are neither politically neutral nor morally consistent; they align with US and allied state interests.
• Some say people should direct anger at lawmakers and foreign policy, not only at platforms.

Social Media Harms and Regulation Ideas

• Many liken big social platforms to tobacco:
  • Addictive by design, optimizing outrage and division.
  • Large societal externalities: polarization, mental health, manipulation, propaganda.
• Proposals:
  • Higher or targeted taxes on ad revenue or “net negative” companies.
  • Treat platforms as publishers once they algorithmically curate feeds (Section 230/product-liability angle).
  • Stronger privacy and child-protection laws; even banning algorithmic engagement optimization or “social networks” for minors.
• Others warn broad bans or “anti-psyop” laws could be abused by governments to suppress dissent.

Alternatives and Individual Responses

• Suggested user responses:
  • Quit Meta products entirely; use direct communication (SMS, calls, in-person) and smaller communities.
  • Move to federated / open platforms (Mastodon, Friendica), or group tools (Signal, Discord, Slack).
• Skeptics note network effects: “social is where the people are,” so leaving can mean losing weaker ties.

Meta Discussion About HN and Language

• Side threads debate the headline’s wording (“Arabia” vs “Saudi Arabia”) and HN’s 80-character title limit.
• Several commenters lament perceived decline in HN comment quality and increasing polarization.

Jurisdiction & Global Reach

• Many debate how a New York court can order domain takedowns worldwide.
• Some call it “performance art,” but others note:
  • ICANN and root DNS are US-based, giving leverage even over country TLDs.
  • Mutual legal assistance treaties and trade agreements can pressure foreign entities.
  • Historic examples: pressure on Sweden over The Pirate Bay; Assange extradition.
• Concerns that stronger enforcement (e.g., via RIPE, US transit providers) could fragment the global internet if IP ranges/ASNs become political tools.

Effectiveness of Takedowns

• Consensus that new domains and mirrors will appear; compared to The Pirate Bay “hydra.”
• DNS-focused injunctions break links (e.g., from Wikipedia) but don’t erase content.
• Suggestions: Tor/onion services for stronger censorship resistance; prediction that enforcement may push AA in that direction.

AI Companies vs Shadow Libraries

• Thread highlights that publishers cited AA as an AI training hub (Meta, NVIDIA).
• AA reportedly offered high-speed bulk access for large donations; at least one major US AI company allegedly paid for it.
• Contrast drawn with big AI firms:
  • They face lawsuits and large settlements, but keep domains and operations.
  • Argument that rich companies can “pay to proceed,” while AA operators risk prison and therefore avoid court.
• Debate over legality:
  • Some say training is treated differently from distribution; AA directly distributes copies, AI companies mostly don’t.
  • Others argue models are effectively “IP laundromats” and should be retrained without infringing data.

Piracy, Authors, and Access

• Sharp split:
  • One side: AA harms authors, publishers, booksellers, and even libraries; undermines future work.
  • Other side: AA provides vital access and preservation; law and current copyright terms (~life+decades) are seen as unjust.
• Piracy framed by some as a “service problem”: if DRM-free, reasonably priced ebooks were easily available, demand for AA would drop.

Libraries, Digital Goods & Control

• Discussion of how digital licensing breaks the traditional “first sale” model:
  • Libraries pay per-loan or time-limited licenses instead of owning ebooks.
  • Digital licenses are expensive; publishers keep control and can limit or stop lending.
• Worries about centralized censorship: far easier to choke off access by not renewing digital licenses than by pulling physical books.
• Some advocate personal archives and decentralized, censorship-resistant “shadow library” designs; others stress the broader social role of public libraries.

Performance and Benchmarks

• Some report asm.js SHA-256 implementations outperforming available WebAssembly (Wasm) libraries in browsers, with specific benchmarks showing:
  • In Chrome on Windows: asm.js ≈ 2× faster than Wasm for a given SHA-256 implementation.
  • In Firefox: Wasm ≈ 2× faster than asm.js (with asm optimizations) and ≈ 2× faster than Chrome’s Wasm for that benchmark.
• Others argue that in modern Firefox, asm.js is compiled through the same pipeline as Wasm, so asm.js cannot be intrinsically faster; better or worse codegen and engine differences likely explain gaps.
• Several commenters stress that “X is faster than Y” is highly context-dependent (browser, platform, module, workload).

---

WebCrypto and Hashing Use Cases

• crypto.subtle.digest is praised for speed but:
  • Requires a secure origin.
  • Is async-only.
  • Lacks incremental hashing, making it unsuitable for very large files without full buffering.
• These gaps motivate custom asm.js/Wasm hashing solutions.

---

Wasm vs asm.js: Capabilities and Integration

• Wasm benefits mentioned:
  • Access to SIMD, bulk memory ops, GC types, externref, evolving proposals (memory control, stack switching).
  • Fewer bounds checks and better memory strategies.
• Criticisms / limitations raised:
  • Isolation from JS and web APIs; most web APIs still require JS “shim” calls.
  • No direct zero-copy sharing of arbitrary ArrayBuffers; extra copies needed when moving data into Wasm heaps.
• Clarifications:
  • Strict asm.js also cannot directly call most web APIs or avoid copies; it is closer to Wasm-with-JS-syntax than to general JS.
  • Some think asm.js “can do everything JS can do,” others correct that it’s numerically constrained and cannot handle JS objects/strings directly.

---

Runtime Code Generation and Tooling

• Concern: losing asm.js optimizations hurts dynamic codegen patterns.
• Others note:
  • asm.js remains valid JS; it just won’t get a special fast path.
  • Generating Wasm at runtime is described as straightforward with small helper libraries or encoders in Rust/JS.
• Legacy pain points:
  • Emscripten dropped asm.js support; compiling old asm.js-targeting code with old toolchains is frustrating.
  • Desire for an asm.js → Wasm transpiler; an older asm2wasm tool existed in Binaryen but is now deprecated.

---

NaCl/PNaCl and Historical Trajectory

• Thread revisits NaCl (native machine code sandbox) and PNaCl (LLVM bitcode) as predecessors/alternatives:
  • PNaCl suffered from heavy startup costs and non-standard APIs tied to Chrome.
  • Wasm is seen as a cleaner, CPU-agnostic, multi-implementation standard.
• Some lament an “alternate timeline” where NaCl/PNaCl or a more mature Wasm replaced today’s heavier Electron-style apps.

---

Ecosystem Maturity and Pain Points

• Reported shortcomings of current Wasm ecosystem (especially in browsers):
  • No direct DOM/web-API access without JS glue.
  • Multithreading via Web Workers is cumbersome and header-dependent (COOP/COEP).
  • No zero-copy streaming interfaces; data marshalling overhead with WebGPU and others.
  • Fragmented runtimes and a messy WASI story; async and threading are awkward.
• Others respond that Wasm is evolving slowly but deliberately (e.g., GC, stack switching, interior pointer discussions, components/WASI work).

---

Real-World Apps and Sentiment

• asm.js credited with enabling early “thick” web apps and demos (e.g., large C++ codebases, Unreal Engine in browser, design tools).
• Some see its removal as technically sensible but emotionally nostalgic; others view it as just retiring an obsolete compilation target.
• Mixed feelings about everything moving into the browser: some praise distribution and collaboration benefits; others wish more for native/desktop options.

Context and overall reaction

• Commenters see repeated booing of AI‑praising commencement speeches as a visible youth backlash against how AI is being sold.
• The phrase “the kids are alright” recurs, meaning the younger generation’s instincts might be healthy despite older complaints about “kids these days.”

Student anger and job‑market anxiety

• Many grads feel betrayed: told for years that college → good jobs, now graduating into layoffs, hiring freezes, high debt, and AI hype.
• AI is perceived as directly or indirectly shrinking entry‑level roles, especially for junior engineers and creatives.
• Some argue layoffs are really from “pandemic overhiring” and capital reallocation to AI infra, but others see that as an increasingly thin cover story.

AI as tool vs labor replacement

• One camp: AI is “just a tool” like past tech shifts; skills remain useful, and workers must adapt.
• Counter‑camp: AI is explicitly marketed by executives as a way to cut headcount; the purpose and incentive structure matter more than the abstract tech.
• Analogies used: crop harvesters replacing farmhands, sewing machines, cotton gins; emphasis that who controls the tool decides who benefits.

Capitalism, shareholders, and distribution of gains

• Strong skepticism that “shareholder value” or trickle‑down will benefit grads; most gains expected to accrue to capital, not labor.
• Retirement funds and older asset holders are seen as aligned with executives pushing AI to reduce labor costs, deepening generational inequality.

Arts, music, and cultural impact

• Artists and musicians are depicted as especially hostile: AI is seen as “content slop” that devalues craft and floods channels.
• Some non‑musicians like being able to generate “good enough” personalized songs; others fear trust in new music and livelihoods will erode.

Commencement speeches and tone‑deafness

• Many think commencement should celebrate and inspire, not pitch AI or enumerate crises.
• Praising AI as an inevitable “rocket ship” to success, or telling students to “deal with it,” is widely viewed as condescending and oblivious to their precarity.

Generational and political dynamics

• Younger commenters describe rising hopelessness, anger at “ladder‑pulling” older generations, and growing openness to extreme views.
• Some see anti‑AI sentiment as a rational class response; others worry enemies of the U.S. could exploit youth anti‑AI attitudes.

Pro‑AI and pragmatic views

• A minority argues AI will increase productivity and prosperity long‑term; resistance only hurts those who refuse to learn it.
• Small business owners and some professionals report real productivity gains from AI assistants and coding tools.
• Others call for coupling AI adoption with strong social supports (e.g., basic income, regulation against AI‑justified layoffs).

Use vs rejection of AI

• Several note a possible tension: students may rely on AI for homework while denouncing it at ceremonies.
• Responses say this isn’t hypocrisy: people can resent a tool they feel coerced into using in a rigged system.

Eric Schmidt–specific issues

• Beyond AI, some grads and commenters object to Schmidt personally: past wage‑suppression scandals and reported sexual assault allegations.
• Local reporting (linked in the thread) notes organized efforts to disinvite him and to encourage booing before he spoke, so motivations are mixed (AI + personal/ethical concerns).

Reliability of Google’s AI Overviews

• Many commenters see Google’s AI Overviews as highly unreliable, often extrapolating from a single obscure source or Reddit comment and presenting it as fact.
• The “hot-dog champion” demo is viewed as trivial in itself, but alarming as proof that one blog post can seed authoritative‑sounding AI answers.
• Users report similar index poisoning: fake whale names, scam support numbers, niche technical “facts,” and spoof products being confidently restated by AI systems.
• Some note that AI summaries blur context: turning one user’s dimensions or anecdote into a “typical” or “official” claim.

Manipulation, GEO/AEO, and Spam

• Many characterize this as the new SEO: “Generative/Answer Engine Optimization,” with agencies already selling services to game AI answers.
• Concern that existing playbooks—blog farms, fake reviews, influencer campaigns, hacked sites—now target AI systems instead of classic search.
• Some worry about higher‑stakes manipulation: health supplements, finance/retirement advice, political narratives, and foreign influence operations.

Data Quality, Training, and Curation

• Several argue that training on the whole internet is like citing tabloids; propose curated or “reference” datasets and better fact/opinion bucketing.
• Others counter that refutations are also in the data; the core issue is task design and prompts, not just training.
• There is broad skepticism that large‑scale, human‑curated corpora are economically feasible.

Trust, Sources, and Ranking

• Suggestions include: surfacing source strength, showing when claims rest on a single or obscure source, and building a “2026 PageRank” for trust.
• Others point out this is hard, political, and gameable; any scalar trust score can be exploited, as with backlinks in the past.
• Debate over whether centralized “trusted news” pipelines for LLMs are desirable or dangerously gatekeeping and politicized.

Google’s Incentives and Track Record

• Some say Google solved spam early (PageRank, ML signals) and is failing to apply that knowledge; others argue Google effectively gave up on spam once ads dominated.
• Multiple comments assert that correctness was never Google’s real product; attention and ad revenue are, so quality control lags until reputational damage forces it.

Broader Views on AI and User Responsibility

• Opinions on AI range from “useless garbage” to “transformative for code generation and systems design.”
• Long subthreads debate whether LLMs are “just glorified search” or exhibit emerging reasoning, with no consensus.
• Several emphasize that critical thinking and skepticism remain essential; naive users over‑trust AI where they once understood search results as “just websites.”
• Some users cope by blocking AI widgets, or even fantasizing about poisoning AI training data as a form of resistance or mischief.

Overall Reception

• Strongly positive reaction; many call it an amazing, formative, or “most awesome” site, happy it survived the Flash era.
• Users enjoy exploring subgenres, validating their own expectations, and rekindling interest in old bands.
• Some humor around expectations that “Map of Metal” might be about chemical elements or Apple’s Metal API.

Implementation, History & Tech

• Site originally built in Flash in about 1–2 weeks, later ported to HTML5 “for old times’ sake.”
• Uses OpenSeadragon for the zoomable map; source code is on GitHub.
• Creator mentions YouTube embedding policies changing over time and past issues with being blacklisted for hiding the player.
• There exists a larger, more detailed physical sketchbook version that isn’t online.

UX, Bugs & Mobile

• Multiple users report issues on Firefox and mobile: stuck Black Sabbath pop-up, no obvious way to close player, or music not changing when clicking regions.
• Author frames it as primarily a desktop experience; switching to desktop mode on mobile can help.
• Some UI confusion: clicking the skull starts interaction, not the label text; no search or gazetteer, which several people miss.

Genre Choices & Omissions

• Users debate subgenres: thrash vs speed metal, Swedish death vs melodic death, tech-death era bias, industrial/metalstep-like styles, sludge vs “atmosludge,” and deathgrind vs grindcore.
• Noted omissions or underrepresentation: Katatonia, Agalloch, Alcest, some cores/tech styles, “Thall,” Linkin Park in nu-metal, fantasy/dwarf metal, certain key bands (e.g., Order from Chaos).
• Some praise specific choices (e.g., song selections for hardcore punk, recognition of Neue Deutsche Härte).

Comparisons & Related Projects

• Frequently compared to Ishkur’s Guide to electronic music, Every Noise at Once, Metal Archives, and other music maps/visualizations.
• Another commenter shares a separate large-scale “music map” project.

Nostalgia & Cultural Commentary

• Strong nostalgia for early web/Flash era: experimental, non-monetized, weird personal projects.
• Critique of today’s ad/SEO-driven web and “10 main websites” monoculture.
• Reflections on metal history influences (Hendrix, Sabbath, Judas Priest, etc.) and how narratives have shifted.

Benchmarks, Comparisons, and Marketing

• Multiple commenters criticize Qwen for benchmarking against older competitors (e.g., Opus 4.6) instead of the latest GPT/Claude/Gemini versions, viewing it as marketing/expectation management.
• Some note mismatches between benchmark results and lived experience: Qwen often looks stronger on paper than in day‑to‑day use.
• Others point out benchmarking lag (e.g., new models not yet in public eval suites).

Model Quality vs Frontier Models

• Many see Qwen as “great for open weights,” close to frontier but not equal to top proprietary models.
• Some users report Qwen 3.6 (especially 27B) being good enough to replace mid‑tier models (e.g., Sonnet‑class) for many coding tasks, but still below top‑end (Opus/GPT) on complex work.
• There is anecdotal debate about Anthropic model regressions (4.6 vs 4.7), with conflicting reports and claims of “nerfing” vs harness issues.

Open vs Proprietary & Hosting

• Qwen3.7‑Max is proprietary; “Plus/Max” lines are generally not open weights. People hope for later open releases or 3.7 analogs to 3.6 open models.
• Some want Qwen hosted by US‑based providers (e.g., Fireworks/OpenRouter/Friendli) for compliance and latency; others note 3.6 Plus already appears via some US proxies.

Local Deployment, Hardware, and Performance

• Extensive discussion on running Qwen 3.6 locally:
  • 27B dense vs 35B A3B MoE trade‑offs: dense is smarter but slower; MoE is much faster with slightly lower quality.
  • Quantization choices (Q4/Q5/Q6, K_M vs K_XL) and KV‑cache strategies strongly affect speed and quality.
  • New MTP (multi‑token prediction) variants can roughly double generation speed in some setups.
• Hardware advice spans M‑series Macs (esp. 64–128GB), Strix Halo boxes, RTX 6000, multi‑GPU rigs, and budget cards (3060, P40), with cost ranges from ~$2.5k laptops to $10k+ GPU builds.
• Consensus: local models are slower and more work to tune but provide privacy and predictable costs.

Coding Agents and Tooling

• Several users report good results using Qwen 3.6 (27B or 35B) with coding agents such as pi, Claude Code as a harness, OpenCode, and VS Code integrations.
• Proper harness configuration (large context, “preserve_thinking”, tools) significantly impacts effectiveness.
• For many, Qwen 3.6 is “good enough” to offload a substantial share of coding tasks from paid frontier models.

Hallucinations, Token Efficiency, and Metrics

• Qwen3.7‑Max is highlighted as SOTA on Artificial Analysis’s “non‑hallucination rate” for omniscience, but commenters stress this metric alone is insufficient:
  • A model can avoid hallucinations by refusing to answer.
  • The Omniscience Index (which balances correctness, refusals, and hallucinations) is viewed as more meaningful.
• Token efficiency becomes a major concern:
  • Some Chinese and Nvidia‑branded models are criticized for needing many more tokens to reach similar performance.
  • Gemma 4 is cited as an example of high token efficiency; Qwen and DeepSeek are sometimes described as “chatty.”
  • Users want models that stay close to frontier quality but minimize tokens for cost and latency.

Censorship, Geopolitics, and Trust

• Several commenters won’t use Chinese‑hosted models for corporate or sensitive work, fearing government access and IP exfiltration.
• Others argue US services pose similar surveillance risks; the debate extends to PRISM‑era programs and global intelligence cooperation.
• Concrete examples:
  • Hosted Qwen models reportedly refuse to discuss Tiananmen or Uyghurs, while “decensored” local variants do.
  • Western models are said to provide more detailed answers on some controversial topics but still show alignment/censorship on others.
• Some Europeans feel stuck between distrusting both US and Chinese providers and lacking strong native alternatives.

Economics, Cloud vs Local, and ROI

• Opinions split on whether to rent GPUs (Runpod/Vast) vs buy hardware:
  • Rentals are often priced to pay off hardware in 1–1.5 years; some argue buying is better if usage is sustained.
  • Others note that “other people’s compute” is simpler and avoids large capex but sacrifices privacy and control.
• Individual anecdotes:
  • Heavy users saw $100–200/month SaaS AI bills, which nudged them toward buying high‑RAM laptops to run local Qwen/Gemma.
  • Some compare local‑LLM expenditure to the 2016–2018 crypto GPU wave, warning about hype vs real productivity.

Adoption, Alternatives, and Switching Behavior

• Noticeable migration patterns:
  • Some users report moving from Google Pro (Gemini/Flash) to Qwen/DeepSeek due to pricing and quota limits.
  • Others shifted from Claude (especially dissatisfaction with 4.7) to Kimi, Qwen, or DeepSeek models.
• People see a rough split:
  • Frontier APIs for highest‑stakes or hardest problems.
  • Open or cheaper models (like Qwen 3.x) for everyday coding and planning, often via local or budget hosting.

State of the C standard library

• Many commenters see C’s stdlib, especially string and number functions, as fundamentally unsafe or poorly designed (lack of bounds checking, locale issues, ambiguous errors).
• Others argue the library is weak but acceptable if wrapped; “C is not its standard library,” and serious C projects often build their own safer utility layers.
• There’s regret that C never got a widely adopted “Boost-like” common library or a single dominant package manager, leading to every shop reinventing utilities.

Integer parsing pitfalls in C

• Built-ins like atoi/atol, strtol/strtoul/strtoull, and sscanf are criticized for:
  • Silent truncation / overflow, or using max values (e.g., ULONG_MAX) as sentinels.
  • Accepting negative input for unsigned parses and wrapping instead of erroring.
  • Stopping at first invalid character and returning a partial value (e.g. "123timmy").
  • Legacy behaviors like octal interpretation of leading 0.
• A concrete example shows strtoull on large negative literals yielding small positive numbers by wraparound, which many consider simply “the wrong answer.”

Workarounds and alternatives

• Common patterns: write your own parser, wrap stdlib functions, use return-code-plus-output-parameter APIs, or error via errno, negative codes, or abort.
• Some propose pre-validating with regex or string comparison round-trips, though that’s seen as ugly or inefficient.
• OpenBSD’s strtonum is noted as better but limited (whitespace handling, only signed long long).
• Example custom parsers are shared; even those have subtle UB bugs pointed out (e.g., negating INT64_MIN).

Language design, UB, and portability

• Strong criticism of UB: compilers can legally drop checks (e.g., null checks, overflow) leading to surprising crashes.
• Debate over C’s “portable assembly” role; some argue its flexible integer sizes undermine true portability, others say efficiency justified the design historically.
• One view: standard functions are lexeme scanners optimized for unbounded Unix text streams, not full validators; proper parsing should be a layer above.

Teaching and philosophy

• Anecdotes: courses assigning “parse integers correctly” as a semester-long exercise to expose edge cases.
• Split perspectives: some say the article nitpicks edge cases; others insist correct, unambiguous parsing is a baseline requirement, not perfectionism.