Hacker News, Distilled

Study methodology & context windows

• Some argue the paper’s design is “flawed” because prompts were asked sequentially in one chat, so earlier prompts (e.g., “answer in three words”) bled into later ones.
• Others point out the paper also ran isolated sessions to explicitly test context dependence; using both is seen as the core of the experiment, not a mistake.
• A subset worries the analysis doesn’t clearly separate those conditions, enabling possible cherry‑picking of results.

What “hallucination” means

• Strong disagreement over terminology: “hallucination” vs “bullshitting,” “confabulation,” “bad output,” “misprediction.”
• Critics say “hallucinate” anthropomorphizes systems with no beliefs or awareness and obscures that this is just erroneous output.
• Supporters say the term is now established, intuitively captures confident fabrication, and is useful for non‑experts.
• Several suggest “bullshitting” in the philosophical sense: fluent, confident speech without concern for truth.

Why LLMs hallucinate – and why they work at all

• One camp: LLMs are statistical next‑token generators; hallucinations are the inevitable result of prediction under uncertainty and compressed world knowledge.
• Another camp says this “just autocomplete” framing is technically true but misleading; internal layers appear to build rich feature/world representations and in‑context learning mechanisms.
• Broad agreement: accuracy is high where training data is dense and consensus exists (e.g., popular languages, APIs); errors spike with sparse, fast‑changing, or controversial topics.
• Some argue information‑theoretic and complexity limits mean hallucinations can never be fully eliminated.

Intelligence, world models, and limits

• Ongoing debate on whether LLMs “have” mental/world models or merely model sources and word co‑occurrences.
• Some see emergent capabilities (multimodal reasoning, internal features that track real‑world entities) as steps toward genuine world modeling and even future “minds.”
• Others insist they lack self‑knowledge and epistemology: they don’t know when they don’t know.

Mitigation strategies & tooling

• Proposed mitigations:
  • RAG with explicit grounding and separate factuality checkers.
  • Symbolic logic / theorem‑proving or semantic validators (e.g., for SQL) to catch structural errors.
  • Better calibration and explicit confidence estimates.
  • Tool use (compilers, interpreters, search) and secondary “fact‑check” passes.
  • Careful sampling/logprob control to trade creativity vs reliability.

Societal and usability concerns

• Many are less worried about LLM behavior than about user interpretation and vendor marketing that portrays them as reliable, intelligent agents.
• Concern that people over‑trust confident answers, especially without domain knowledge or visible provenance, leading to misuse and misallocation of resources.

FAA Licensing Delay & Process

• SpaceX says Starship Flight 5 hardware has been ready since early August but the FAA now estimates a late‑November launch license, versus prior expectations of mid‑September.
• The company frames the delay as driven by unnecessary environmental analysis rather than new safety concerns, and criticizes the ability of consultations to reset 60‑day clocks repeatedly.
• Some commenters see this as typical regulatory pacing; others describe the FAA as unusually slow and possibly weaponizing process against SpaceX.

Environmental Impact & Water Discharge

• A major thread is the deluge system: SpaceX emphasizes it uses potable water and claims post‑use samples show contaminant levels below discharge standards.
• Critics counter that:
  • Output water is heated, mixed with exhaust, metals, and pad contaminants.
  • SpaceX lacks proper industrial wastewater permits.
  • You are required to prove safety before discharging, not dump first and test later.
• Supporters reply that:
  • Regulators (state environmental agency, FAA, Fish & Wildlife) reviewed the system and initially judged it acceptable.
  • The key question is actual contaminant levels, not that water is “processed by a rocket.”
• There is disagreement over the seriousness of regulatory violations (minor paperwork vs. substantive Clean Water Act issues).

Role of FAA, NEPA, and Bureaucracy

• Some argue the FAA is the “lead federal agency” coordinating with environmental regulators, which is standard practice and may prevent conflicting demands.
• Others think tying environmental clearance to launch licensing gives the FAA excessive gatekeeping power and amplifies NEPA‑driven delay.
• Broader concern: public‑comment and impact‑statement processes are seen by some as easily gamed by special interest groups to stall projects.

SpaceX’s Track Record and Trust

• Critics point to:
  • Damage from the first Starship launch (pad destruction, debris spread).
  • A pattern of permits pushed to the edge or violated.
  • A combative, unprofessional tone in SpaceX’s press release.
• Defenders argue:
  • Rapid‑iteration R&D inherently accepts early failures.
  • Environmental impacts so far have been limited and mitigated.
  • Oversight should focus on real harm, not paperwork missteps.

Politics, Musk, and Perceived Retaliation

• Some see delays as politically motivated punishment tied to Musk’s right‑leaning, antagonistic online behavior and clashes with the current administration.
• Others note that policy toward Tesla/SpaceX has been materially supportive overall and attribute friction more to union politics and Musk’s personal grievances.
• There is debate over whether individual regulators or rival contractors are “out to get” SpaceX versus simply enforcing rules.

National Security and Space Race Framing

• Pro‑SpaceX voices stress:
  • Starship’s strategic value for cheap mass‑to‑orbit, Artemis, and competition with China.
  • The risk that excessive regulation undermines US space leadership.
• Skeptical voices question:
  • Whether a few‑month slip materially affects strategy.
  • The wisdom of granting a single private company outsized influence over critical space and defense infrastructure.

Privacy and Surveillance Concerns

• Many see in-cabin microphones and ad-targeting as equivalent to installing a “bug” in the car.
• Cars are perceived as semi-private spaces, especially for conversations; audio monitoring feels more invasive than cameras.
• People are uneasy about audio being sent to remote servers, likening this to smart speakers and smartphones, but note that many consumers already accept such devices.
• Some suggest physical countermeasures (cutting mics/antennas, jamming with white noise), and want right-to-repair to explicitly protect this.

Ownership, Subscriptions, and Ads

• This is framed as another step toward not truly owning cars, but renting a surveilled service.
• Heated-seat subscriptions and other “features as a service” are cited as precedent; many expect ads without real price reductions or opt-out options.
• Comparisons are made to mobile games and streaming: “watch ads for credits,” enshittification, and dynamic pricing based on conditions and user profile.
• Some predict initial deployment in rental fleets where ad revenue can subsidize purchase price.

Market Dynamics and Consumer Choice

• One camp believes people can just avoid Ford; another argues all automakers will copy profitable surveillance and consumers generally don’t prioritize privacy.
• Difficulty of buying “non‑smart” TVs and phones is used as an analogy for where cars may end up.
• There’s discussion of choice architecture and how markets present constrained, pre-structured options, limiting real agency.

Legal and Regulatory Questions

• Commenters speculate about conflicts with one‑party vs. all‑party consent and expectations of privacy inside cars.
• Some note existing practices: telematics, OnStar, Nissan-style terms requiring owners to “inform occupants,” and startup EULAs used as blanket consent.
• It’s unclear how aggressively prosecutors or regulators will enforce wiretap or privacy laws against such systems.

Patents and Corporate Intent

• Some argue patents don’t guarantee implementation and might even block competitors; others see the filing as a clear signal of desired monetization.
• The patent text explicitly acknowledges users’ desire for “minimal or no ads” and describes optimizing ad load for revenue, which many read as confirming hostile intent.
• Broader frustration is expressed with the patent system, corporate greed, and weak democratic checks on such practices.

Car Tech Trends and “Dumb Cars”

• Several want “dumb cars” with minimal electronics and plan to keep older vehicles; others think such cars and manual driving will be regulated away over time.
• There’s mixed sentiment on touchscreens: some praise newer models retaining physical controls; others note consumer demand for features like CarPlay drives screens, even if UX suffers.

Pricing and Value vs. PCs

• Many see $699 (without disc drive) as too close to capable PC builds; some claim a ~$900 PC can beat or match it, especially when factoring PS Plus fees and cheaper PC games.
• Others argue modern GPUs are so expensive that even midrange PC builds far exceed the PS5 Pro’s total cost, so $700 is “what it costs now.”
• Several note that last gen “Pro” consoles launched at or near base-console prices, whereas this is a sizable step up with no corresponding base-model price drop.

Hardware, Performance, and Use Cases

• Anticipated benefits: higher resolutions, more stable 4K, better frame rates, improved upscaling, and headroom for titles like GTA 6.
• Skeptics say current games rarely max out the base PS5; many expect 30 fps caps to remain in heavier titles, making the upgrade feel marginal.
• Console convenience (no tinkering, stable platform, couch play, good HDR) remains a major draw versus PC headaches (drivers, crashes, configs).

Physical Media, Digital-Only, and Ownership

• Strong pushback against the disc-less default and costly add‑on drive.
• Arguments for discs: cheaper new and used prices vs. monopoly digital stores, resale and lending, library and sharing within families, long‑term access, and Blu‑ray playback.
• Concerns that newer PS5 revisions require online activation for drives, weakening preservation and offline use.
• Some PC‑first players say they went fully digital long ago and don’t miss discs.

Platform Strategies and Market Positioning

• Sony is criticized for chasing high-end hardware and subscription revenue while lacking a deep, evergreen first‑party IP stable compared to Nintendo.
• Microsoft is viewed as pivoting toward services and multi‑platform releases; some see Xbox hardware as increasingly secondary.
• Nintendo is repeatedly praised for a “sane” strategy: affordable, portable hardware anchored by strong exclusive IP, even on outdated tech.

Target Audience and Upgrade Incentives

• Many question who this is for: too expensive as a family gift, but not compelling enough for enthusiasts already on PC.
• Some expect it to sell mainly to graphics/FPS obsessives and late adopters coming from PS4, not existing PS5 owners.

Header handling strategy & risks

• Many are surprised Cloudflare uses a “list of header names that are internal” instead of structural separation or strict prefixes.
• Concerns raised: name collisions with user headers, inconsistent lists across services, sanitization bugs, and issues with Connection header semantics.
• Some argue this pattern is common in large enterprises and edge proxies; others say that doesn’t make it less fragile.
• Several suggest prefixing all internal headers (CFInt, X-CF-) and stripping by prefix, but others note legacy systems, early headers, third‑party appliances, and acquisitions make global renaming hard.
• It’s stated that a longer‑term plan is to stop using HTTP headers for internal IPC entirely; some worry that makes the trie work a short‑lived stopgap.

Alternative designs proposed

• Separate metadata channel: dedicated internal protocol (e.g., Protobufs or custom encapsulation) instead of overloading HTTP headers.
• Structural approaches: maintain a list of “allowed to exit” headers instead of “internal to strip” (deny‑by‑default), or record original inbound headers and only emit those.
• Data‑structure tweaks: force internal headers to the front and remove first N; use a header-count sentinel; or tag headers as internal at creation rather than inferring later.

Tries vs hashes vs regex

• Discussion on why a trie beats hash tables here: hashing strings requires touching every byte; tries often reject on the first character, and most lookups are misses.
• Alternatives floated: custom fast hash functions, perfect hashing, Bloom/binary-fuse filters, hardware CRC32, or specialized hash maps; others point out these still need substantial hashing work.
• Regex/Aho‑Corasick and DFAs are mentioned as conceptually similar; regex libraries carry general-purpose overhead, DFAs can be faster but use more memory and build time.
• Some critique the article’s Big‑O characterizations, arguing they blur key factors like cache behavior versus comparison counts.

Performance impact & ROI debate

• The function optimized is in an extremely hot path, and small per‑request wins aggregate across tens of millions of requests per second.
• Some see saving hundreds of cores as modest vs overall fleet and question engineering ROI compared to tackling larger architectural issues.
• Others counter that recurring CPU, power, and capacity savings, plus improved headroom, justify micro‑optimizations in hot code, and that the write‑up also has marketing and educational value.

Hardware evolution: GPUs, ASICs, and specialization

• Many argue current “GPUs” like H100 are already quasi-ASICs: graphics features stripped, heavily optimized for matrix multiplications.
• Some expect further specialization: transformer- or even model-specific ASICs, possibly with weights on-chip, enabling huge tokens/sec but less flexibility.
• Others warn transformer-specific silicon is risky: architectures (Mamba, RWKV, hybrids) are changing fast; ASIC cycles are too long, favoring GPGPU/TPU/NPU-style flexibility.
• Debate over programmability: some say inference (and even training) doesn’t need general programmability; others insist future algorithms will.

Will GPU demand collapse, plateau, or keep rising?

• Critics of the headline say “demand ends” is wrong; at most, growth may slow or shift from new to used hardware.
• One view: if GPU performance per dollar keeps improving and task demand stays flat, unit sales could fall sharply.
• Counterview: demand is not fixed; cheaper compute opens new applications (historical parallel with CPUs and software).

Brute force vs smarter algorithms

• Many agree we’re in a brute-force phase: giant models, huge datasets, lots of parallel matmul.
• Some think improved algorithms will reduce reliance on massive GPU fleets and could obsolete today’s specialized hardware.
• Others see brute force as inherent to ML (hyperparameter sweeps, large search spaces) and expect compute appetite to persist even with better methods.

Use cases and “we’re just getting started”

• Optimists claim we’re early: text, images, and audio are “checked”; video, 3D, simulation, planning, reasoning, etc. are still emerging and will require far more compute.
• Predicted future uses include: high-fidelity video generation, holodeck-like XR experiences, pervasive small-scale LLMs in products, richer game NPCs, and expanded computer vision workloads.
• Skeptics counter that generative quality gains are slowing, many applications are gimmicky, and previous tech hype cycles (crypto, metaverse, self-driving) overpromised.

Economics, efficiency, and induced demand

• Strong thread on Jevons paradox: greater efficiency often increases total resource use (example analogies: roads, developer tools, historical CPU gains).
• Others stress limits: energy costs, thermals, and opportunity cost mean not all idle compute should be used; some hardware (old GPUs/CPUs) becomes uneconomic to run.
• Unclear how these forces net out, but most agree parallel compute remains fundamentally valuable beyond the current LLM boom.

Overview of ruling

• ECJ upheld the European Commission’s view that Ireland granted Apple unlawful state aid via bespoke tax rulings, and ordered Ireland to recover ~€13B plus compound interest.
• Most commenters stress this is not a “fine” but back taxes that should have been paid under EU state‑aid rules; no additional penalty was imposed.

EU law, sovereignty, and retroactivity

• One camp argues EU law has primacy in areas like state aid and the single market, so national tax rulings can be invalidated even decades later.
• Others push back, citing national constitutional limits, veto powers, and examples where courts in countries like Poland or Germany have resisted full EU primacy.
• Debate over whether this is “retroactive” law: critics say the arm’s‑length principle and interpretation weren’t clearly in force; defenders say the underlying treaty rule (no selective state aid) existed since the 1970s and was merely enforced late.

Why Apple pays vs Ireland

• Some object that Ireland made the bad deal, so Ireland should be punished.
• Replies: the formal decision is against Ireland, which must now recover illegal aid from its beneficiary, Apple; companies can’t rely on unlawful promises.
• Several note the perverse optics: Ireland enjoyed jobs and investment, then receives the back taxes, though at the cost of reputational damage and tighter future scrutiny.

Ireland’s tax strategy and fairness within EU

• Many see Ireland as a de facto tax haven that “hacked” the single market, undermined other members’ tax bases, and advantaged US multinationals over EU firms.
• Counter‑arguments: Ireland was historically poor, used low corporate tax and English language/common law to attract FDI, and other states also run favorable regimes (e.g., patent boxes, sectoral subsidies, Netherlands/Luxembourg structures).

State aid vs normal tax competition

• Core legal distinction: low general rates and published schemes (available to all firms that meet criteria) are broadly allowed; secret, bespoke rulings for single firms are not.
• Commenters emphasize Apple’s ultra‑low effective rate (~0.005%) as clear evidence of selectivity.

Impact on business, innovation, and citizens

• Some believe the sum is manageable for Apple but a significant signal that aggressive tax planning in the EU is risky.
• Others worry about legal uncertainty, business hostility, and broader EU over‑regulation harming tech innovation.
• For Ireland, commenters highlight potential use for infrastructure (e.g., metro) but note existing budget surpluses and capacity/planning, not cash, as main bottlenecks.

Overall reception

• Many commenters are impressed by radiology-specific performance, especially on formal exams and radiograph tasks.
• Others reserve judgment until third‑party validation and real‑world deployment data are available.
• Some note this is far from the first “AI for radiology” effort; the commercial path is seen as the real difficulty.

---

Benchmarks, exams, and claims

• Reported results on the FRCR 2B Rapids mock exam (radiographs only) are seen as strong; some ask if the model was trained on exam questions.
• A developer states the model was not trained on FRCR questions and clarifies it only took a mock rapid‑reporting component, not the full exam, and that this has since been clarified on the site.
• Comparisons to general multimodal LLMs (GPT‑4o, Gemini, Claude, etc.) are questioned because most weren’t trained specifically on diagnostic imaging.

---

Access, openness, and datasets

• Several people can’t find a public model or code; access appears gated via a waitlist and eventual commercialization.
• Commenters wish for an open, “LLaMA‑style” radiology foundation model; current open efforts are mostly narrow (e.g., lung cancer) or small research models.
• Suggestions for datasets include TCGA/NCIA, DeepLesion, MIMIC CXR, and commercial vendors.

---

Clinical context and workflow

• Radiologists stress that diagnosis is not pure image classification; patient demographics, history, and symptoms matter.
• The model’s use of both images and chart data is praised as more realistic.
• Multiple comments complain bitterly about RIS/PACS/EMR fragmentation and messy metadata; integration and data quality are seen as a larger barrier than model accuracy.
• There is strong interest in tools that:
  • Auto‑structure dictations and reports.
  • Explain specific image regions with literature references.
  • Triage studies and reduce radiologist burnout.

---

Ethics, public access, and economics

• One camp argues that restricting public access is largely greed; another cites safety, misdiagnosis, and system strain from “confidently wrong” self‑diagnosers.
• Debate over whether access limits are “infantilization” vs necessary stewardship, with examples from different countries’ pharma and diagnostic access.
• Concerns that AI may become another billable line item while justifying staff cuts and shifting liability.
• Radiology remains highly competitive as a specialty; several argue AI is more likely to augment overworked radiologists than replace them soon.

Google’s Role in the Web Ecosystem

• Many argue Chrome and Google’s dominance in web standards lets them protect ad and platform monopolies, weaken privacy, and avoid commoditization of Android APIs.
• Others counter that Google “saved” the web, pushed PWAs, and enabled sophisticated web apps (e.g., Figma‑like) versus a world dominated by locked-down app stores.
• There’s nostalgia for a Firefox‑led ecosystem, but recognition that Mozilla is financially dependent on Google.

Third‑Party Cookies and Conflicts of Interest

• Several comments say third‑party cookies likely persisted mainly because Google feared antitrust fallout from killing them.
• Mixed views: technically useful for some benign cases, but widely abused and mostly unwanted by privacy‑minded users.
• Consensus that having the dominant browser vendor also be a top ad‑tech player creates deep conflicts of interest.

Ad‑Tech “Trifecta” and Market Structure

• Core allegation discussed: Google dominates three layers of display ads—publisher ad server, ad exchange, and advertiser tools (“sell side,” “auction,” “buy side”).
• Commenters describe Google as simultaneously exchange, broker, market‑maker, and a big advertiser, with insider data and control over auction rules.
• Examples raised include Jedi Blue (coordination with Facebook) and Project Bernanke (tweaking auctions to benefit Google’s side), though details and interpretations are disputed.
• Some ad‑tech practitioners stress the product maze is confusing even for insiders and may be intentionally so.

Effects on Publishers, Newspapers, and Advertisers

• DOJ narrative (as relayed in comments): Google’s position lets it both underpay publishers and overcharge advertisers, skimming large margins.
• Some say this likely worsens the economics for news sites already hit by the loss of classifieds and changing reader behavior.
• Others are skeptical that Google alone “killed newspapers,” pointing instead to Craigslist, the open Internet, and weak demand for paid journalism.

Antitrust Standards, Outcomes, and Comparisons

• Discussion contrasts U.S. “consumer welfare” (price‑focused) with broader European dominance standards; some see both as ill-suited to “free” ad‑funded services.
• Debate over whether breaking up monopolies is usually good: many cite Standard Oil, AT&T, airlines; others note mixed or negative cases (rail, energy, post‑Soviet transitions).
• Some think this is the strongest antitrust case against Google (clear market structure, measurable economic harm); others doubt courts will meaningfully act, citing the Microsoft case.

Apple, Meta, Amazon, and Market Power

• Repeated comparisons:
  • Apple—seen by some as a worse mobile gatekeeper (App Store control, 30% cut), but with separate ongoing DOJ actions.
  • Meta—huge in social ads, but perceived as less monopolistic today in social media overall.
  • Amazon—viewed as offering great consumer value while still potentially abusing platform power (especially via AWS and marketplace).
• Some worry that weakening Google on Android could unintentionally strengthen Apple’s mobile dominance in the U.S.

Privacy, Surveillance, and Broader Harms of Ads

• Several express frustration that the case targets competition and publisher harm, not privacy abuses or societal damage from surveillance advertising and attention‑driven media.
• Others reply that antitrust law is about economic harm, and privacy needs separate legislative frameworks.
• There’s a minority view that advertisers mostly want cohorts and distributions, not granular personal data, while others insist advertisers do aggressively seek detailed profiles.

Complexity, Obfuscation, and Public Understanding

• Commenters note that Google’s internal guidance discourages defining “markets” or “market share,” which is seen as defensive lawyering.
• Many argue the technical and contractual complexity of ad‑tech itself serves to hide anticompetitive behavior and make effective regulation harder.
• Overall sentiment: the ad‑tech stack is opaque, misaligned with user interests, and any remedy will be technically and politically messy.

Overall reaction & legacy

• Commenters express strong sadness despite his advanced age, saying his death feels like losing part of their own past.
• Widely described as a “true legend” whose presence elevated even modest or campy material.
• Many highlight how central his work was to their childhoods and cultural memory.

Iconic roles & performances

• Frequently cited roles: Darth Vader (Star Wars), Mufasa (The Lion King), Thulsa Doom (Conan the Barbarian), Admiral Greer (Jack Ryan films), Terrence Mann (Field of Dreams), King Jaffe Joffer (Coming to America), roles in Dr. Strangelove, The Hunt for Red October, Sneakers, The Sandlot, Matewan, Cry the Beloved Country, Command & Conquer: Tiberian Sun, Mathnet/Square One, Reading Rainbow, and The Simpsons appearances.
• Some emphasize that Conan and Field of Dreams, not Vader, are their primary associations.
• Several note the irony that in some of his best roles he barely uses his “infamous voice” or plays against type.

Voice, narration & readings

• His voice is repeatedly called uniquely authoritative and impossible to replace.
• Mentioned narrations and readings include: Edgar Allan Poe’s “The Raven,” Frederick Douglass’s “The Meaning of July Fourth for the Negro,” the Bible, university hype videos, GPS and telecom ads, and children’s shows.
• Comparisons are drawn with other distinctive voices (e.g., Morgan Freeman, Dennis Haysbert, various veteran voice actors), but many feel no younger actor matches his gravitas.

Career, background & craft

• Noted biographical points from the thread: childhood stutter overcome via reciting poetry, Ranger School completion in the 1950s, early film work in Dr. Strangelove.
• Several recall his own advice about staying a “journeyman” and never treating any role as the final word.

Personal encounters & impact

• Multiple posters share in-person stories: kindness to a stuttering child in a recording booth, generous backstage conversations after Othello, friendly gestures on a film set.
• These anecdotes reinforce an image of him as gracious, patient, and down‑to‑earth.

Debates, corrections & meta

• Disagreement over Star Wars billing: some argue he was under-credited; others note he initially declined credit out of respect for the on‑screen actor, and that later editions credit him.
• A claim that Dr. Strangelove inspired the White House Situation Room is corrected with timeline details and a note about a president later confusing the film set with reality.
• Brief meta-discussion on whether his passing warrants special treatment on Hacker News and whether it counts as “tech news.”

What Cohost Was & Why Some Liked It

• Described as a small, non-enshittified social network with a pleasant UI and strong moderation norms.
• Users liked: chronological feed; no global “firehose”; no like/repost/follow counts; long-form post culture; HTML/CSS posts enabling creative layouts; cohesive repost/comment threading.
• For some, it felt less “cramped” and less cesspool-like than Twitter/X, Facebook, or even Mastodon.
• Others bounced off it: small network, not compelling enough vs. existing apps, unclear why to use it over alternatives.

Centralization vs Federation

• Some saw lack of federation and account portability as a deal-breaker, making shutdown inevitable and offramp painful.
• Counterargument: federation wouldn’t fix revenue, would add large engineering and moderation burdens, and complicate culture control.
• Decentralized services are seen as messy (defederation, instance politics, migration), which turns some users off.

Costs, Salaries, and Sustainability

• Many surprised at ~$41k/month in expenses for a small userbase; discussion concludes most of it is payroll and benefits for four under-market-rate employees.
• Debate over whether $90k-ish pay is “high” or simply necessary to live and build a serious product.
• Thread repeatedly contrasts real labor costs (dev, admin, moderation) vs users’ focus on cheap VPS hosting.

Funding, Shutdown Mechanics, and Stripe

• Cohost’s codebase is collateral on a loan, blocking open-sourcing; speculation that lender may just delete it or sell it.
• Stripe reportedly killed a key tipping/subscription feature that didn’t gate exclusive content, hurting revenue.
• Some are baffled by loan terms; others frame this as a typical venture-debt/financing risk.

Ethical Social Media & Business Models

• “Ethical” is variously defined as: no surveillance capitalism, no ad-driven outrage optimization, user-serving features.
• Skeptics argue “ethical social media” is either a niche hobby or a money pit; without profit, it can’t scale to mainstream.
• Others counter that profit motive is harmful and that large-scale social media should resemble a public utility, possibly publicly funded.

Broader Implications & Emotional Reactions

• Cohost’s failure raises worries about long-term viability of Bluesky, Tumblr, Mastodon/Fediverse, etc., once investor money ends.
• Some see Mastodon/fediverse growth as modest and possibly a “dead end”; others note it “works” for niche communities.
• Multiple commenters are saddened; users describe losing a supportive community and creative space, with no guarantee of a comparable replacement.

Examples of Software-Driven Hardware Degradation

• Numerous anecdotes: soundbars, smart speakers, monitors, TVs, VR headsets, sous-vide devices, printers, smart coffeemakers, consoles (PS3 OtherOS), Firesticks, smart TVs, Synology NAS apps, car infotainment systems with injected ads.
• Common patterns: forced updates that brick or remove features; paywalls or “activation” fees added post-purchase; remote service shutdowns that kill basic functionality.
• Mixed experiences with support: some vendors replace bricked hardware; others demand costly RMAs or simply abandon products.

Cloud Dependence and “Smart” Devices

• Strong resentment of devices that require vendor servers for essential functions (garage doors, IoT appliances, cameras, robot vacuums).
• Concern about products becoming useless when cloud services shut down or business models change.
• Some users now refuse to connect appliances or buy “smart” anything, or segregate IoT on VLANs with minimal/zero WAN access.

Proposed Legal / Regulatory Approaches

• Mandate that cloud-tethered products work locally and retain at least original functionality for a fixed period (e.g., 7–10 years).
• If a company reduces features, shuts servers, or EOLs products, require:
  • Refunds/compensation based on depreciation; or
  • Release of firmware, server code, APIs, keys, and docs into public domain or open source.
• Ideas to treat software sunsets like right-to-repair, or like patent expiration.
• Some see this as anti-fraud/anti–bait-and-switch; others frame it as antitrust against bundling hardware with captive services.

Open Firmware and Interoperability

• Calls for legally required unlockable bootloaders and sufficient hardware documentation.
• Suggestions to remove DMCA protections on firmware/drivers to enable community maintenance.
• Acknowledge obstacles: third‑party proprietary components, code shared across active product lines.

Certification, Labeling, and Disclosure

• Proposals for FTC-backed labels: open source, cloud‑free, telemetry‑free, E2EE, firmware rollback, long-term support/parts.
• Comparisons to nutrition or energy-efficiency labels; debate over effectiveness vs. regulatory capture and consumer confusion.
• Some argue such products should be clearly marked as “dependent on manufacturer servers” and potentially “revocable.”

Skepticism and Counterpoints

• Free‑market view: don’t buy such products; competition will fill the gap.
• Others argue information asymmetry, lock‑in, and industry-wide practices make “just don’t buy” unrealistic.
• Questions about FTC authority after recent court rulings, and fear of leadership changes reversing progress.
• Practical worries: enforcing obligations when companies go bankrupt; feasibility of large-scale open-sourcing; reliance on mobile platforms that themselves deprecate apps/APIs.

User Workarounds and Alternatives

• Adoption of local‑first ecosystems (Zigbee, Z‑Wave, Matter/Thread, Home Assistant, Tasmota, ESPhome, Valetudo, Shelly).
• Preference for “dumb” appliances, older cars without locked-down software, and avoiding cloud features entirely.
• Some admit resorting to ethically dubious returns/swaps as a response to perceived planned obsolescence.

Upgrade behavior and device longevity

• Many are upgrading from iPhone 7–12 era devices; 11/12/13 Pro owners are split on upgrading vs waiting another cycle.
• A lot of people happily use 4–8‑year‑old phones; main forced-upgrade drivers are app support drops, battery degradation, and network changes (4G→5G).
• Several mention “trickle‑down” within families: newest phone bought yearly, older phones handed down.

Form factor, displays, and ergonomics

• Strong frustration that Pro models got larger/heavier again; small‑phone fans (SE / 12–13 mini) feel abandoned.
• Many say 60 Hz on non‑Pro models is unacceptable in 2024; others insist most people can’t see or don’t care about 120 Hz.
• OLED vs LCD is contentious: some love OLED blacks; others dislike PWM flicker, low‑brightness behavior, and burn‑in risk.

Camera and new camera button

• Camera upgrades (48 MP wide, 5× telephoto, 4K120, external ProRes recording) are widely seen as the only clearly compelling change, especially for parents and casual creators.
• Dedicated capacitive camera button is liked in principle (half‑press focus, zoom slider, mode shortcuts), but some note past attempts (Sony) saw limited adoption.

Apple Intelligence and AI integration

• Reactions split: some see video/photo semantic search and device‑aware Siri as “obvious next step” that will soon feel indispensable; others see gimmicks (stickers, rewrites) and fear spammy tie‑ins (Apple TV+ promotion, notifications).
• Concern that many AI features are delayed, US‑only at first, English‑only at launch, and possibly constrained in EU due to regulation.
• Privacy skeptics distrust opaque on‑device/cloud mix; others contrast Apple favorably with data‑hungry competitors.

Battery, repairability, and environment

• Battery life gains on 16/16 Pro are appreciated, but lack of user‑replaceable batteries is repeatedly criticized as anti‑environmental and profit‑driven.
• Apple’s official battery replacement is seen as technically available but too expensive/inconvenient compared to a hypothetical quick-swap design.

Pricing, storage, and positioning

• Keeping 128 GB as base is called “stingy,” especially with growing photo/video sizes; others say cloud storage makes it fine for most.
• Pro vs non‑Pro gap feels smaller this year: key differentiators are 120 Hz, materials, LiDAR, telephoto zoom, USB 3, and durability.
• Overall sentiment: technically solid but “boring” update in a mature, plateaued smartphone category; many will only upgrade when their current phone fails.

Overall reception

• Many commenters find Series 10 iterative rather than groundbreaking.
• Some are glad there’s no Ultra 3 this year (no upgrade/FOMO), others disappointed by lack of major advances (especially battery and sensors).

Health features & sensors

• Blood oxygen measurement is removed on US-sold units due to patent issues; some consider buying abroad to keep it.
• Rumored passive blood pressure and blood glucose did not appear; several say the tech isn’t ready yet.
• New sleep apnea detection is seen as potentially useful, though some doubt its value for “borderline” cases. It reportedly uses motion rather than SpO₂.
• Training load / readiness features are welcomed, especially by people who struggle to interpret body signals; comparisons made to Garmin and Whoop-style metrics.

Battery life & charging

• 18-hour “all-day” claim is widely criticized, especially for sleep tracking, long hikes, ultras, and travel.
• Some say fast charging plus a daily routine (e.g., during shower/coffee) makes battery a non-issue; others reject needing “charging habits” at all.
• Ultra’s 36–72 hour life is praised but its price and size are barriers. Garmin and other devices with multi‑day or multi‑week battery are frequently cited as superior on this axis.

Design, size & materials

• Thinner case is appreciated by some (less bulky under sleeves), but many would prefer more battery over thinness.
• Others want smaller, more elegant, or circular options; some find Apple Watch visually bland compared to traditional watches.
• New titanium and glossy black finishes generate interest, but there is concern about scratching; sapphire on premium models is praised and users complain it remains gated from base aluminum.
• Larger Series 10 display is attractive, especially for aging eyes; some analysis suggests it’s now larger than Ultra 2’s actual display area.

Ecosystem, independence & connectivity

• Persistent frustration that Watch setup and full use still require an iPhone; users want pairing with Mac/iPad or true standalone operation.
• Some run nearly phone‑free on LTE watches and like that, but others say many apps still assume the phone is nearby.
• Cellular add‑on pricing from carriers is seen as high; MVNO options are discussed with caveats about deprioritization and roaming.

Alternatives & competition

• Garmin is repeatedly recommended for serious fitness, battery life, and outdoor use, though weaker as a “general smartwatch” and more fragmented in product line.
• Whoop is discussed as a no‑screen, subscription tracker with strong behavioral impact for some, but the subscription cost is widely disliked.
• Pebble, Amazfit, Xiaomi bands, Pixel Watch, and classic Casios surface as options for those prioritizing simplicity and long battery.

Scope of the New Feature & Device Support

• Feature targets H2-chip devices; some expect it on AirPods 4, others note Apple only named AirPods Pro 2 in the FDA context, so broader support is unclear.
• It’s being positioned as a regulated “medical device” use, with explicit mention of ongoing FDA and other-agency approvals.
• Commenters expect AirPods Pro 2 to remain on sale for many years if they’re part of a medical device line.

How It Compares to Traditional Hearing Aids

• Many report clinical hearing aids are very expensive (often thousands of dollars) and have poor music/Bluetooth quality versus AirPods/Bose/etc.
• Lifespan estimates vary: some say ~3 years for “features like Bluetooth,” others indicate devices can last longer but become outdated.
• Multiple people emphasize that proper hearing aids do complex, frequency-specific processing tailored to an audiogram; simply “turning up the volume” is not enough.
• Experts in the thread say Apple’s solution is for mild–moderate loss and the “gap years” before full hearing aids, not a replacement for cochlear implants or serious loss.

AirPods Fit, Comfort & Daily Use

• Strong divide on fit: some say AirPods Pro stay in through running; others say they constantly fall out regardless of stock tip size.
• Many recommend third‑party tips (foam, wings, sleeves), custom molds, or hooks; foam tips improve isolation but degrade faster and can affect charging.
• Wearing AirPods all day is considered difficult or uncomfortable by some, limiting their use as a full-time hearing aid.

Existing iOS Hearing Features & Apps

• Several note that similar functionality already exists via:
  • iOS Headphone Accommodations + audiogram import.
  • Mimi hearing test app to generate an audiogram and sync it to Apple Health, then systemwide.
  • Conversation Boost on AirPods Pro for speech-in-noise enhancement.
• Some report added processing modes can introduce latency and reduced quality compared to standard transparency.

Regulation, Market Impact & Cost Disruption

• The new OTC hearing-aid category and deregulation are credited with enabling such products and lowering prices.
• Hearing-aid industry is described as under-innovative, insular, and overpriced; Apple and others (e.g., Sony, Jabra, Sennheiser’s hearing-aid ties) are seen as potential disruptors.
• Even with frequent AirPods replacement, some argue total cost can still undercut traditional aids.

Stigma, Accessibility & Social Dynamics

• Multiple comments highlight stigma around visible hearing aids, especially for younger people; AirPods could normalize assistive hearing tech and serve as an “on-ramp.”
• Others note a reverse issue: wearing AirPods can signal “I’m not listening,” which may be problematic in social or school settings.
• Several users with borderline or situational hearing difficulty see this as a compelling middle ground vs. committing to multi‑thousand‑dollar aids.

Limitations, Safety, and Reliability Concerns

• Some stress that hearing aids are medical devices with strict safety, robustness, and lifecycle requirements; concern exists about Apple potentially discontinuing support.
• Others counter that workplaces should not rely solely on hearing for safety, but commenters with severe loss describe real risks if aids fail.
• Battery life and wearing comfort are seen as key blockers to treating AirPods as full-time aids.

Other Observations & Open Questions

• Questions remain about:
  • Whether the new mode is “clinical-grade” transparency or something more advanced.
  • All-day (12+ hour) battery options and support for features like CROS (single-sided deafness).
  • Android-compatible alternatives; the thread notes there are many new OTC options but does not converge on a specific best choice.
• Some mention difficulty trying AirPods in-store; others suggest returns within 14 days or sanitized demo units in certain regions.

Device Reliability and Study Design

• Some worry AirPods may lose high‑frequency output over time, risking misleading hearing tests if Apple relies on them.
• Questions raised about whether AirPods can self‑diagnose frequency range; expectation that Apple will still advise professional hearing tests.

AirPods, ANC, and Tinnitus

• Multiple anecdotes that newer AirPods (especially Pro Gen 2, Max, HomePods) trigger or worsen tinnitus at relatively low volumes, compared to other headphones/speakers that feel safe even when loud.
• Others emphasize that AirPods track exposure, show volume alerts, and that ANC can allow listening at lower volumes, potentially reducing risk.
• Theories mentioned:
  • ANC/in‑ear designs may either contribute to or help prevent tinnitus; no consensus or hard evidence in the thread.
  • Fit/pressure and resulting muscle tension in jaw/neck/around the ear can “trigger” tinnitus independent of volume.

Psychological and Perceptual Aspects

• Large subthread on habituation: focusing on tinnitus amplifies it; distraction, acceptance, and not constantly researching it often reduce distress.
• Many report that when deeply engaged (e.g., games, audiobooks, work), they don’t notice it; quiet rooms make it more intrusive.
• Tinnitus often correlates with stress, fatigue, anxiety, and high blood pressure; managing these seems to reduce perceived loudness for some.
• Several describe tinnitus as an “information/attention hazard”: once mentioned, people notice their own more.

Causes and Comorbidities

• Reported triggers include loud noise (e.g., music, drumming), scuba/pressure issues, infections, sinus and eustachian problems, posture/neck tension, TMJ, gum/root infections, Viagra and other vasodilators, and hearing loss itself.
• Some have pulsatile tinnitus tied to heartbeat; commenters urge medical evaluation because it can indicate other conditions.

Treatments, Coping, and Aids

• Hearing aids often reduce tinnitus by restoring missing frequencies/background noise; Apple’s new AirPods hearing‑aid‑like features are seen as potentially disruptive to the traditional, expensive hearing‑aid market.
• Non‑device strategies: posture correction, neck/jaw massage and stretching, magnesium supplementation (anecdotal), relaxation/meditation, brown/white noise or music at night, audiobooks, and specific sound therapies (e.g., ACRN‑style tones).
• A few mention drugs (e.g., memantine, SSRIs) as possibly reducing severity but note side‑effects and the need for short, cautious use.
• Overall consensus: protect hearing, get evaluated for underlying issues, and manage the psychological response as much as the sound itself.

Hardware & Feature Segmentation

• Debate over why sleep apnea detection is limited to Apple Watch Series 9/10 (and not even Ultra 1).
• One side argues it’s largely artificial segmentation, similar to Apple Intelligence being limited to newer iPhones despite cloud processing.
• Others point to plausible technical reasons: newer accelerometers, need for additional sensors (SpO2, possibly temperature), higher compute, and battery constraints.
• Some accept Apple’s pattern of withholding technically-possible features if they degrade user experience (speed, battery).

Sensors, Patents & FDA Status

• Disagreement over whether apnea detection uses only the accelerometer or also blood oxygen and temperature.
• US models have SpO2 disabled due to a patent dispute; non-US specs still list it, suggesting hardware is present but blocked.
• Speculation that Apple could use optical data “internally” without exposing explicit SpO2 readings.
• Feature is not yet live; commenters say it’s pending FDA clearance.

Accuracy & Role of Watch-Based Detection

• Dedicated PAT-based devices and full sleep studies use multiple channels (SpO2, ECG, airflow, belts, video) and even then can be ambiguous.
• Many view watch detection as a screening tool akin to AFib alerts, not a definitive diagnosis.
• Concerns that intermittent SpO2 sampling on watches may miss short desaturations, though long-term nightly data could still be useful.

Oxygen Monitoring & Battery Life

• Some wish for continuous nocturnal SpO2, but others cite battery and regulatory limits.
• Apple aggressively duty-cycles sensors to hit ~18‑hour life; critics argue Apple over-prioritizes screen/CPU over multi-day health tracking.
• Users share charging strategies (during shower, coffee, commute) and note better life on Ultra models.

CPAP Access, Regulation & Market Structure

• Strong frustration with gatekeeping: prescriptions, insurance delays, repeated studies, and cost.
• Some advocate OTC, mass-market CPAPs with user-adjustable pressure, arguing risks of self-titration are lower than untreated apnea.
• Others stress the need for regulation and physician oversight, citing complexity of correct use and examples like Philips’ foam recall and CPAP-related deaths.
• Side discussion on at-home tests, online vendors, and using hidden “service menus” to tweak settings.

Alternative Remedies & Fringe Ideas

• Mouth taping and a 19th‑century “shut your mouth” book spark heavy skepticism; critics emphasize lack of modern evidence and safety concerns.
• Another commenter’s claim that briefly increasing head blood pressure can “100% fix” apnea and myopia is widely challenged as implausible.
• Cheap off-brand watches with SpO2 are warned against due to reports of obviously fake readings.

Kernel architecture and “hybrid” vs monolithic

• Debate over whether “hybrid kernel” is meaningful: several argue NT is effectively monolithic today, with many components (including graphics) in kernel mode.
• Some note NT 3.x started more “microkernel‑ish” (user‑mode graphics/print) but performance concerns pulled much of that into the kernel in NT 4 and later.
• Comparisons to Darwin/Mach: user perception is “just UNIX”, but implementation (Mach + BSD) is quite different from classic Unix.

Processes, threads, and async I/O

• Strong focus on NT’s “threads are the schedulable unit; processes are containers” model vs traditional Unix process‑centric culture.
• Several posters clarify that modern Unix/Linux kernels also schedule threads, and that Linux “process vs thread” is mostly policy over the same task structure.
• NT’s async I/O (OVERLAPPED I/O, IOCP, Registered I/O, IoRing) is praised as integrated and general (files, sockets, etc.), with kernel‑level asynchrony from the start.
• On Unix, POSIX AIO and assorted event mechanisms (kqueue, Solaris ports, epoll) are seen as more bolted‑on and inconsistent; io_uring is viewed as the first Linux facility comparable to NT’s best APIs but has had security issues.

Filesystem and performance differences

• Repeated claims that Windows filesystem operations (especially many small files) are slower than Linux; others counter that raw FS performance is similar and overhead comes from filters like Windows Defender.
• Discussion of DevDrive as a mitigation (reduced filtering) and of NTFS’s highly flexible, IRP‑based stack vs Linux’s more special‑cased, optimized paths.
• File locking semantics and “file in use” behavior on Windows are blamed for some performance and usability pain.

Heritage: VMS, RSX, MICA

• Strong consensus that NT’s real lineage is DEC’s RSX‑11 → VMS → MICA, not Unix.
• Many architectural features (AST/APC, IRPs, object manager, SMP design) and even the “WNT = VMS+1” naming joke are cited as evidence.
• Some note that decades of accreted features and compatibility have made NT much more complex than its clean original design.

Security, objects, and registry vs /etc

• NT’s uniform object model (handles, security descriptors, signaled state, unified wait) is widely admired; contrasted with Unix’s bolt‑on object and ACL concept.
• Some argue NT’s security architecture is theoretically stronger; others point out that user‑mode cruft and third‑party software often bypass or weaken those guarantees.
• The Windows registry is compared to /etc as a hierarchical key–value store; critics dislike its opacity, GUID forests, and uninstall cruft, while defenders say the concept is fine and problems are mostly developer practice.

CLI, arguments, and Unicode

• Windows’ “single command line string + per‑program parsing” is seen by some as a serious design flaw (inconsistent quoting, no canonical Argv→string transform).
• Others argue Unix shell‑side globbing is itself dangerous and archaic (cp * issues, filenames starting with -), preferring Windows‑style program‑level expansion.
• Long subthread on UTF‑16/UCS‑2 vs UTF‑8: NT’s early Unicode adoption is praised; its legacy UCS‑2 roots and surrogate handling are criticized. Many now see UTF‑8 as the more practical default, but timing made NT’s choice understandable.

WSL, subsystems, and POSIX layers

• Historical POSIX subsystems (Interix, Services for Unix) and OS/2 subsystems are recalled as architecturally elegant but ultimately abandoned.
• WSL1 is admired as a true NT subsystem with syscall translation; disappointment is common that WSL2 reverted to VM‑based Linux for compatibility and performance.
• Some speculate NT could have evolved into a stronger multi‑personality kernel; others note business and maintenance realities.

Backwards compatibility and bloat

• Many praise NT/Windows for extraordinary backward compatibility (16‑bit apps via NTVDM on 32‑bit, decades‑old binaries still running).
• Others worry this locks in bad design decisions and bugs that can’t be fixed, and complicates security hardening.
• UI and shell: strong nostalgia for Windows 2000/7‑style dense, “classic” desktop; current Windows 10/11 UX is criticized as sluggish, over‑animated, and trend‑driven, though some note LTSC + tweaks can still feel very fast.

Why GitHub Won vs. Other Forges

• Many see GitHub’s success as a mix of timing and “taste”: launched as DVCSs were rising, focused narrowly on Git, and shipped a clean, developer-centric UX when others were clunky or ad-driven.
• Pull requests, easy forking, and a pleasant web UI greatly lowered the barrier to contributing to open source, versus mailing-list patches and centralized commit rights.
• Competitors (SourceForge, Google Code, Bitbucket, CollabNet) are described as enterprise- or distribution-focused, slow to adopt Git, or lacking polish and consistent UX.
• Network effects reinforced the lead: once “all the cool projects” were on GitHub, people migrated even if they preferred Mercurial or other tools.

Git vs. Other Version Control Systems

• Older developers recall CVS/SVN/Perforce/Visual SourceSafe as painful, especially for branching, merging, offline work, and server dependence.
• Git’s local commits, cheap branching, and offline workflows were a huge leap, even if its CLI UX is considered confusing and “user-hostile.”
• Several argue Mercurial or bzr had better UX and concepts, but lost out due to performance, weaker hosting ecosystems, and GitHub’s Git-only bet.
• Some note that most teams now use Git in a de facto centralized way, with GitHub as “the server,” blurring DVCS vs. centralized distinctions.

SourceForge, Google Code, and Monoculture

• SourceForge is remembered as initially crucial, then increasingly “enshittified”: ad-saturated pages, misleading download buttons, and eventually bundled adware/“grayware.”
• A Google Code insider states its goal was to break SourceForge’s monoculture, not to “win” or make money; they shut it down once alternatives existed, keeping an archive.
• Others criticize this posture as a cop-out and cite the shutdown as part of a broader pattern of Google abandoning products users rely on.

Concerns About GitHub Dominance and Lock-In

• Some worry GitHub is now its own monoculture, but point out viable alternatives (GitLab, Bitbucket, self-hosted Gitea/Forgejo, SourceHut, etc.) and easy migration of raw Git history.
• Lock-in is seen more in ecosystem features (issues, PRs, cross-repo references, GitHub Actions) than in Git itself.
• There is discomfort with a proprietary, Microsoft-owned platform sitting at the center of open source, especially given Copilot’s training on GitHub code.

Project overview

• Project Hammer compiles historical Canadian grocery prices from major chains and releases them publicly (e.g., SQLite dump) to enable academic and legal analysis and, ideally, increase competition and deter collusion.
• Some commenters are enthusiastic about the transparency and potential for consumer tools (price trackers, comparison apps, community buying, etc.).

Collusion vs. competition

• Several point to Canada’s proven bread price‑fixing scandal and coordinated “hero pay” and seasonal price freezes as evidence that grocers can and do collude.
• Others stress that similar or synchronized prices can arise from healthy competition, common suppliers, price‑matching, or shared cost shocks; correlation alone is not proof of collusion.
• There is concern that highly transparent, real‑time price data can actually facilitate tacit collusion: firms track each other and quickly match price hikes instead of undercutting.

Evidence and international parallels

• Examples cited from Austria and Norway show near‑simultaneous, identical price moves across chains; some see this as damning, others attribute it to shared suppliers or signaling.
• Norwegian and Austrian cases illustrate both explicit collusion findings and suspicion where proof is lacking.

Limits of price‑data analysis

• Antitrust practitioners in the thread argue that detecting collusion from price series alone is extremely hard; robust cases usually require internal documents, communications, or “motel‑meeting” evidence.
• Grocery markets have thousands of products, regional pricing, promotions, and volatile input costs, increasing false‑positive risk if one just hunts for suspicious patterns.
• They encourage treating the dataset as a “hammer” that enables hypothesis generation and public scrutiny, not as a standalone detector of illegality.

Canadian market structure & broader causes

• Many characterize Canada as oligopoly‑heavy (grocers, telecom, banks), with weak competition policy, frequent consolidation, and some protectionist or supply‑management regimes (dairy, poultry).
• Debate over food inflation: some emphasize rising gross margins and corporate profits; others point to monetary policy, fuel costs, war, COVID disruptions, and concentrated suppliers as primary drivers.
• There is broad agreement that grocers are part of the problem, but disagreement on how large a part compared with macro policy and upstream consolidation.