Hacker News, Distilled

McDonald’s app & trusting the client

• Multiple commenters are surprised this is yet another case of McDonald’s apps lacking proper server-side validation, despite wide publicity.
• Core critique: heavy anti-reverse-engineering and root checks give a false sense of security and encourage developers to “trust the client,” which is fundamentally flawed.
• Some see the impact as mostly McDonald’s losing money and reputation; others worry about precedent and what else the same teams might build.

Root detection, DRM, and device control

• Many criticize Play Integrity / SafetyNet–style checks and root detection: they don’t stop serious attackers but punish power users and reduce device ownership.
• Several banking and corporate apps are cited as refusing to run on rooted phones or even phones with sideloaded apps; other commenters say those same apps run fine for them, suggesting OS/version differences or inconsistent checks.
• Some defend these checks as risk management and tech-support reduction, but others see them mainly as liability-shifting and control.

Android/iOS sandboxing, filesystem, and backup

• Discussion on Android’s changing storage model: older versions allowed broad filesystem access; newer ones sandbox more tightly, improving security but complicating backups.
• Users lament the lack of a robust, system-level, app-data backup interface and see Google nudging people toward its cloud sync.
• iOS is viewed as more locked down but also less leaky in terms of apps inspecting the device.

Developer incentives, outsourcing, and security culture

• Several argue that outsourced/mobile “app mill” work optimizes for shipping quickly and passing checklists, not real security.
• One contractor openly describes doing the bare minimum and waiting to bill fixes later, prompting pushback around ethics and client risk.
• Some note that security “best practices” often function more as liability cover than true protection.

Unions, professionalization, and labor market

• A thread explores unions or professional bodies for developers to refuse insecure or unethical work.
• Others doubt this is realistic: accreditation brings questions about who is blamed in a team and whether members would actually be protected.
• Mixed views on the job market: easy in some EU countries for mid+ roles; difficult for seniors and in the US since recent layoffs.

Other examples & user experiences

• Examples include insecure ticketing/public transport apps, heavily obfuscated IoT apps (e.g., Tuya ecosystem) that fight local control, and anti-cheat in games installing kernel-level components.
• Polish McDonald’s users complain about declining coupon value, clunky kiosks, and being forced into the app for decent prices.
• Some refuse to install such apps at all or leave 1-star reviews when root checks block them.

Devin and agentic workflows: strengths

• When tasks are well-scoped, stacks are mainstream, and tests are easy to run, Devin can produce clean, test‑passing code and handle multi-file changes autonomously.
• The “agent in a Slack/terminal” UX and closed-loop workflows (edit → run tests → iterate) impressed many and shifted expectations about what’s possible.
• Some see current results as analogous to early image generation: rough now, but the mere fact it works at all suggests large future upside.

Major limitations and failure modes

• Tends to make extraneous edits beyond the request, sometimes breaking unrelated functionality, and is bad at rolling those changes back.
• Often gets stuck in “infinite thinking loops,” working for hours or days instead of asking for help, especially on “soft stops.”
• Poor at admitting incapacity or integrating coaching; frequently compared to the worst stereotype of an overconfident junior dev.
• Subtle, hard‑to‑spot mistakes (e.g., silently truncating a license header) undermine trust.
• Users report no reliable way to predict which tasks it will succeed on, limiting its value as a tool.

Agents vs. narrower tools

• Many argue Devin overreaches; narrower agents focused on bug fixes, small features, test/CI cleanup, or maintenance show much higher success rates and real enterprise interest.
• Constrained agents and IDE‑integrated tools (Cursor, Copilot, Aider, OpenHands, others) are seen as more practical: they act as “power tools,” not replacements.
• There’s discussion of orchestrators, time/“energy” limits, and supervisors (even non‑LLM models) to detect when an agent is stuck and halt or escalate.

Where LLM coding helps today

• Explaining legacy or “arcane” code, proposing refactors, and writing tests with many edge cases.
• Generating small, next‑step snippets in data science, SQL, matplotlib, shell/HTTP work, and onboarding to new technologies.
• Automating tedious chores: merge conflicts, linter fixes, parameter reshuffling, multi‑file refactors.

Models, hallucinations, and trajectory

• Some report newer reasoning models (e.g., o1 variants, Claude Sonnet 3.5) hallucinate less for coding with good prompts and short context; others still feel “burned” and prefer to write code themselves.
• Debate over whether progress is still rapid or already hitting diminishing returns.
• Broad agreement that AI cannot yet replace engineers; its output needs review comparable to a brand‑new hire.
• Many expect continued pressure from companies to cut headcount using AI, with disagreement on how far that will actually go.

Cryptography Techniques for Private AI

• Discussion of homomorphic encryption (FHE) and MPC:
  • In principle, FHE can support neural network operations, but current implementations (especially gate-level binary FHE) are ~10⁶× slower than plaintext.
  • CKKS-style schemes are more practical for ML: ResNet-20 inference can be done in minutes on CPU, with hopes of ~1s on small networks using hardware acceleration.
  • Large models like LLMs remain “unreasonably slow” under FHE for the foreseeable future.
• MPC and libraries like CrypTen can hide user inputs from the model owner, but outputs are still visible to the provider.

End-to-End Encryption vs. User Agency

• E2EE protects data in transit but often coincides with poor or nonexistent data export features, limiting user control and portability.
• Some see this as deliberate lock-in; others argue it’s more about lacking incentives to build good export tooling.
• Moving accounts (e.g., device-to-device transfers) is not the same as users having raw, scriptable access to their own encrypted data.

Apple’s Private Cloud Compute and Confidential Computing

• Many view Apple’s PCC / secure enclave approach as a pragmatic, privacy-improving step compared to standard cloud AI.
• Others stress that PCC is still just a technical guarantee: it can reduce insider and attacker access, but does not inherently provide transparency or limit secondary use of data.
• There is mention of Nvidia H100 and cloud GPU enclaves (Azure, possibly AWS/GCP) being used to build similar “encrypted-to-enclave” AI services.
• Some participants argue the article overstates the need for cloud inference, noting Apple Intelligence is restricted to devices powerful enough to run models locally.

Surveillance, Policy, and “Who the AI Works For”

• Strong concern that AI plus cloud services will enable mass, automated surveillance and “thoughtcrime” detection:
  • Existing trends: content scanning for CSAM, extremist threats, “grooming,” drugs/sex/guns, protest/union organization.
  • Worry that LLMs easily normalize slang, coded speech, and embeddings can be inverted to recover text.
• Fears that AI-based detection systems will:
  • Have high-stakes false positives with poor human recourse.
  • Be used for censorship, political repression, or automated law enforcement.
  • Become “accountability sinks” that let institutions blame opaque models.

Cloud AI Business Models and Incentives

• Multiple comments argue incentives, not technical limits, are central:
  • Ad- and data-driven models push providers to scan and retain user data.
  • “Free” or subsidized AI features create lock-in and recurring subscriptions.
  • Without strong regulation and transparency, AI agents are expected to serve providers, advertisers, and governments more than users.

Scope of the FTC Action

• Many see the settlement as narrow: GM is barred 5 years from selling/ sharing individualized driving data, but can still share “anonymous”/aggregated data.
• Some argue this is mainly punishment for deceptive enrollment and lack of consent (OnStar “Smart Driver” dark patterns, hidden sign‑ups, secret insurance profiles).
• Others say if the practice is harmful it should be banned permanently, not time‑limited.
• Debate whether the FTC “cares about privacy” vs. just policing “unfair or deceptive practices” within its limited mandate.

Anonymized vs. Aggregated Data

• Strong skepticism that “anonymous” data is actually safe:
  • Location traces can be re‑identified with a handful of data points or cross‑referenced with app/cell data.
  • Aggregation has also been de‑anonymized in past cases.
• Several commenters argue you must assume the worst: any collected data can eventually be tied back to individuals.

Broader Privacy and Data Rights

• Repeated calls for a US equivalent of GDPR or a “Bill of Data Rights”:
  • People want ownership/control of their data and compensation if it is monetized.
  • Frustration that US law heavily favors data exploitation and “making money” over privacy.
• Comparisons with Europe:
  • Under GDPR, this sort of sale of personal driving data would likely be unlawful without explicit, informed, revocable consent and erasure rights.
  • Some commenters detail how GDPR requires unambiguous, freely given consent and easy withdrawal, and note that deceptive consent flows are non‑compliant in theory, but enforcement is uneven.

Consumer Defenses and Practical Workarounds

• Strategies discussed:
  • Buy older or simpler cars with no telematics; keep repairing them.
  • Research how to disconnect or remove telematics/LTE modules (examples given for specific models).
  • Worry that mandated SIM/eCall and future emissions telemetry in newer cars will make true opt‑out impossible.
  • Concern that leasing or “connected services” apps effectively force data sharing (e.g., loss of remote start).

Systemic Concerns

• Fear of surveillance being used beyond ads: insurance pricing, denial of coverage, law enforcement, courts.
• View that companies keep years of profit and face only light, delayed penalties.
• Split sentiment: some welcome the FTC action as progress; others see it as a weak “yellow card” that leaves the underlying surveillance business model intact.

Motivations and Concerns Around Passwordless TPM FDE

• Many find passwordless TPM-based full-disk encryption (FDE) inherently risky under physical-compromise threat models.
• Others argue it’s “good enough” for opportunistic theft: a random thief is unlikely to exploit OS or bootloader vulnerabilities.
• Several people see automatic encryption as dangerous UX: users can be locked out by hardware failure or policy without realizing their data was ever encrypted.

Use Cases: Laptops, Servers, and Remote Systems

• Laptops: Many prefer typing a strong passphrase at boot; some use biometrics for OS login but still want FDE with a password.
• Home/remote servers and NAS: Owners want protection if disks/servers are stolen but also require unattended reboot, making manual passphrases impractical.
• Corporate environments: Compliance teams often mandate TPM/Secure Boot based on vendor messaging, not deep understanding.

Alternatives for Unattended Boot

• Network-based unlock: Clevis/Tang, dracut-sshd, Dropbear, Mandos allow remote or network-bound decryption.
• Hardware tokens/USB: Random keys on USB or FIDO2 tokens used at boot, with plausible deniability and convenience, but risks of loss/theft.
• kexec tricks: Boot into a small environment that forwards the FDE passphrase to the real system.

TPM, PCRs, and Proposed Mitigations

• Discussion of sealing LUKS keys to TPM PCRs (0,2,7,8,9,15) and how updates break measurements.
• The specific attack works even if kernel/initrd are measured, because it swaps the encrypted volume and relies on the initrd’s password fallback.
• Mitigations proposed:
  • Measure or verify the encrypted volume itself (e.g., LUKS header hash, “prehash” file on rootfs).
  • Use PCR15 as a one-shot: extend it with random data after first unseal so later code can’t re-unseal.
  • Combine LUKS encryption (confidentiality) with dm-verity/fs-verity (integrity).
  • Use intrusion switches or case tamper detection to lock TPM.

Secure Boot, initramfs, and OS Ecosystem

• Criticism that desktop Linux typically does not sign or verify initramfs, weakening Secure Boot.
• UKI and tools like openSUSE’s sdbootutil aim to bind kernel+initrd+command line to TPM measurements and update them automatically, but workflows are complex.
• ChromeOS, Android, and Apple are cited as examples where verified boot plus encryption and a smaller, simpler boot chain provide stronger guarantees.

Disk Encryption Tradeoffs and Threat Models

• FDE simplifies RMA/disposal and allows fast data destruction by wiping TPM keys.
• Physical destruction of drives is debated as simpler vs wasteful and not always reliable.
• Concerns about future cryptanalysis mostly target public-key schemes; symmetric AES with long keys is seen as safer.
• Some argue per-file or per-user encryption (Android, APFS) is more user-friendly than classic FDE, but desktop support is limited.

Windows/BitLocker and TPM+PIN/Password

• BitLocker is discussed; it also faces TPM-bypass style issues in other research.
• TPM+PIN is viewed favorably but rarely deployed; combining TPM with a user password (deriving/wrapping keys) is suggested as a stronger hybrid.

Overall framing

• Most commenters reject the idea that “the world” is becoming uninsurable; they argue specific regions and risks are becoming uneconomic to insure at past prices.
• “Uninsurable” in practice usually means: the actuarially fair premium is either illegal (due to caps) or politically impossible for most customers to pay.

---

Insurance economics and correlated catastrophes

• Insurers must cover expected losses plus a modest margin; for highly correlated events (wildfire, hurricanes, floods) they need years of profit to fund rare, very bad years.
• When risk rises (more frequent fires, higher rebuild costs, denser development), required premiums rise sharply; people accustomed to low premiums perceive this as “gouging.”
• Some note that many P&C and health insurers run on low single-digit net margins; the big dollars flow more to providers, pharma, and occasionally to integrated conglomerates.

---

Regulation, price caps, and market exit

• In California and Florida, commenters point to:
  • Rate caps and slow approval processes.
  • Restrictions on using catastrophe models or reinsurance costs in pricing.
  • Litigation-friendly environments (especially FL).
• Result: insurers limit exposure or leave; “insurer of last resort” pools (e.g., FAIR) grow, often underpriced, implicitly socializing future losses onto taxpayers or other policyholders.
• Several argue price controls are politically popular but ultimately force shortages and hidden subsidies.

---

Climate change vs. development and building standards

• One camp stresses climate change: warmer seas, more extreme heat/drought, and more billion‑dollar events are raising physical risk.
• Skeptics counter with data suggesting no clear long‑term trend in hurricane frequency/intensity, attributing rising losses to:
  • More and pricier assets in harm’s way.
  • Suppression of controlled burns and poor forest management.
  • Building sprawling, flammable suburbs in wildland–urban interfaces and floodplains.
• Broad agreement that:
  • Fire‑ and wind‑resistant construction (concrete/ICF, stucco or fiber‑cement siding, Class A roofs, ember‑proof vents, defensible space) works but is underused.
  • Legacy housing stock and zoning make rapid retrofits difficult.

---

Fairness, subsidies, and “managed retreat”

• One side: living on coasts, in canyons, or in floodplains is a choice; others inland shouldn’t subsidize repeated rebuilds of high-end homes.
• The other side highlights:
  • Long‑standing communities (often poorer or redlined) now facing climate‑amplified risks with little ability to move.
  • Transaction and financing costs (high rates, sunk mortgages) that trap owners.
• Proposed responses include: risk‑based premiums with no caps, stricter building codes, buyouts with no‑rebuild clauses, and ultimately “managed retreat” from some areas.

Firebase billing behavior and “surprise bills”

• Firebase budgets are alerts only; they do not stop usage. Several commenters only realized this from the docs after seeing this incident.
• The official “avoid surprise bills” guidance focuses on alerts and monitoring, not hard enforcement, which many feel does little to actually prevent surprises.
• Free tiers can be hard‑limited or throttled, but once on paid plans, enforcement becomes softer and more opaque.

Hard billing caps: desirability vs difficulty

• Many want an opt‑in hard cap: on hitting a limit, services would be throttled or shut off until manual re‑enablement.
• Supporters argue even an approximate cutoff (with some overrun) is far better than a 3–4‑order‑of‑magnitude surprise.
• Others argue robust caps are technically very hard across many services:
  • Some usage (e.g., network egress) is known only with delay.
  • Enforcing per‑request checks would add latency or require complex distributed coordination.
  • “At rest” resources (storage, backups) raise questions about what to stop or delete.
• Quotas, max autoscaling, and per‑resource limits are proposed as partial mitigations.

Security, misuse, and Firebase complexity

• Firebase is easy to start but “hard to master.” Misunderstood Firestore rules and billing per document processed can cause huge costs.
• It’s easy to accidentally expose data or allow malicious writes, which can generate runaway usage.
• The thread attributes this incident to storing around a petabyte in GCS in a day, possibly via a poorly controlled implementation.

Risk management: LLCs, insurance, cards

• Some advocate using LLCs per app or for hosting contracts to contain liability; others note courts can pierce the veil for fraud, negligence, or “sham” entities.
• Ideas for insurance against cloud overages are floated but widely doubted as practical or fairly priced.
• Virtual cards with low limits help contain card charges but do not erase legal liability for the debt.

Alternatives and broader incentives

• Several prefer fixed‑price VPS or simpler clouds (traditional droplets, Lightsail‑like products, Fly.io with prepaid credits, Supabase, etc.) for predictable bills.
• Some see providers’ refusal to offer caps as primarily profit‑driven; others emphasize enterprise customers’ preference for never‑down services and argue refunds for edge cases are the de facto safety valve.

AI‑generated code, understanding, and professionalism

• Many comments latch onto the prediction that a project will discover large amounts of AI-generated code whose “authors” don’t understand it.
• Strong consensus that submitting code you cannot explain is unprofessional, regardless of whether it came from an LLM, StackOverflow, or elsewhere.
• Some argue this can be a fireable offense, especially where security or confidentiality are at stake; others see it as a coaching opportunity for juniors.
• Reviewers say they would not reject code solely for being AI-generated, but they expect the submitter to explain behavior, correctness, and implications.

LLMs vs StackOverflow and learning practices

• Several people note that blindly pasting from StackOverflow has been a long-standing problem; LLMs mainly amplify this.
• Differences highlighted: LLMs can generate larger, integrated chunks of code and adapt to the user’s context, increasing the temptation to skip understanding; SO answers at least have visible peer review and require some integration effort.
• Some see LLMs as excellent for translation, boilerplate, and “rubber-ducking,” while warning they hallucinate APIs, mis-handle edge cases, and are weak on newer or niche libraries.
• Concern about “learning debt”: juniors and students may advance by outsourcing thinking to LLMs, only to hit a wall later when deeper understanding is suddenly required.

Organizational controls: review, tooling, and risk

• Experiences differ widely: some teams have tight CI/static analysis and block “funky” code; others deploy to production within an hour with minimal review.
• Static analysis and quality gates are seen as helpful but not sufficient; they can enforce style and catch trivial issues, but not guarantee design quality.
• Some worry about maintainer burnout from low-quality “drive‑by” LLM PRs in open source.

Open-source funding and Linux ecosystem

• Debate over whether Linux and key libraries are dangerously underfunded.
• One side claims critical software is maintained by “hobbyists” and that large organizations and governments should fund it at scale.
• Others counter that most kernel work is already done by paid professionals, but acknowledge many crucial user‑space tools and libraries remain single‑maintainer, volunteer‑driven and thus risky.

Free/“ethical” LLMs and copyright

• Some want “truly free” models that do not rely on mass, unpaid scraping of copyrighted material; others argue current copyright law is too restrictive and that broad training use should be allowed.
• There is concern that small players lack the legal cover large companies have when training on potentially infringing data.

Security, maintainers, and geopolitical risk

• Single‑maintainer projects are discussed as both a liability (bus factor, coercion risk) and, paradoxically, simpler to trust because there’s one known person to evaluate.
• XZ-style backdoors are expected to recur; some speculate such attacks might be quietly monetized rather than disclosed.
• Geopolitical fragmentation is seen as a growing risk, though there is disagreement on how much it will actually disrupt open source collaboration.

Other technical notes

• Brief mention of Rust-for-Linux continuing despite a high-profile maintainer’s resignation.
• sched_ext is noted as promising, with at least one concrete gaming-related scheduler example.
• Concerns are raised about cloud‑tied hardware being bricked when vendors fail or shut services, reinforcing “you don’t really own it” worries.

Project overview & initial reactions

• Repository solves the first 100 Project Euler problems, each in a different language.
• Many commenters find the project impressive and fun, especially the breadth of languages and the author’s per-language notes.
• Some expected “10,000 solutions” (100 problems × 100 languages) based on the title and were briefly disappointed before realizing that would be unreasonable.

---

Language selection and omissions

• Interest in more “unusual” choices, e.g., Verilog or other hardware description languages; one commenter notes Verilog is in fact included.
• Others highlight the use of multiple BASIC dialects and note that BASIC variants can be quite different.
• Debate over “proper” statically typed languages:
  • One commenter criticizes the list for omitting popular statically typed languages (e.g., Rust, C++, Go, etc.).
  • Others point out many static languages are present (e.g., Ada, Nim, Haskell, TypeScript, Verilog), and accuse critics of not reading the list carefully.
• Rust’s absence is discussed:
  • Some joke that it “feels like a statement.”
  • It’s clarified the author chose languages that were new to them.
  • Side-thread devolves into criticism of Rust’s community “cult-like” enthusiasm vs defenders framing this as normal tool advocacy.

---

Impressions of specific languages

• Nim receives especially positive commentary: easy to start with, fast compilation, small binaries, pleasant syntax, good for hobby projects.
• Some discuss BASIC, assembly, and BCD (binary-coded decimal) as “old school” concepts that surprised readers when they appeared.

---

Math vs programming in Project Euler

• Several comments stress Euler as primarily math- and algorithm-focused, more so than LeetCode or Advent of Code.
• Later problems often require mathematical insight (number theory, generating functions, closed forms) before coding.
• Suggested resources include classic number theory texts and generatingfunctionology.
• Some argue that brute force becomes unwelcome at a certain point; others say it’s fine to just start and learn as you go.

---

Publishing solutions & respect for Project Euler

• Debate over whether publishing solutions is disrespectful:
  • One side finds public solutions annoying and believes it trivializes others’ efforts and undermines the “honor system.”
  • Others argue:
    • Solutions to early problems are already widely available.
    • The Euler site explicitly permits discussion of the first 100 problems, provided the focus is on methods, not just answers.
    • Project Euler is not a competition; copying solutions is distinct from truly solving problems.

Launch objectives & vehicle changes

• Flight 7 used “Starship v2” upper stage: ~2m longer, redesigned propulsion for ~25% more propellant, modified flap placement to reduce heating, new heat-shield tiles, and non‑structural “catch pins” to study future tower catches.
• Planned profile: booster return and chopstick catch near launch site; upper stage near-orbital trajectory, deploy Starlink simulators, engine relight test, and controlled splashdown near Australia.

Outcome: booster catch vs upper stage loss

• Booster: tower “chopsticks” catch succeeded again, widely viewed as a major engineering milestone showing repeatability, with praise for the precision control and surreal visuals.
• Ship: telemetry showed abnormal behavior (one engine out, possible leak indicated by faster methane depletion, visible flame near flap hinge) before loss of comms and a dramatic high‑altitude breakup over the Caribbean.
• SpaceX’s postflight comment (relayed in-thread): likely oxygen/fuel leak in a cavity above the engine firewall, over-pressurizing beyond vent capacity; planned mitigations include better leak checks, fire suppression in that volume, more vent area. Next launch is not expected to slip far.

Safety, debris, and aviation impacts

• Multiple ground and in‑flight videos showed a striking debris plume; commenters called it beautiful but also unsettling, evoking memories of past crewed disasters.
• Discussion over whether the automated Flight Termination System triggered; later posts say SpaceX indicated it simply exploded.
• Commercial flights near the Caribbean diverted or held, with ATC audio showing extra caution due to uncertain debris paths. Some reports of lightweight debris (e.g., tiles) reaching downrange beaches.
• Thread references NOTAM/NOTMAR maps, international space liability rules, and US insurance requirements; consensus is risk to people is very low but non‑zero and tightly regulated.

Streaming, platforms, and scams

• Many watched via independent YouTube channels; debate about one channel’s NASA‑adjacent branding.
• Complaints about SpaceX using X instead of YouTube: casting/ads issues, lack of official YT feed creating room for scam “Elon/crypto” mirror streams on hijacked channels.
• Suggested mitigations: stricter live‑stream gating by platforms, better channel security (hardware keys), more aggressive takedowns.

Value, timelines, and philosophy

• Enthusiasts emphasize: rapid iteration, unprecedented scale, reusability, and emotional impact as reasons this matters for cheap access to space and long‑term human survival.
• Skeptics counter: Starship is years behind its own Mars/Artemis timelines, has yet to deliver real orbital payloads, and failures create delays and aviation disruption; some see this as over‑marketed and enabling a misleading CEO narrative.
• Comparisons with Falcon 9, SLS, New Glenn, Saturn V, Shuttle, and earlier vertical‑landing experiments fuel debate over whether SpaceX is uniquely “fast” or just loudly iterative.

Origins of Language vs. Writing

• Several comments stress the difference between language (an evolved human capacity) and writing (a recent technology).
• Consensus in the thread: spoken language predates writing by a very long time, likely as long as anatomically modern humans have existed.
• The actual origin of language is labeled “basically unknown”; cave art and storytelling are suggested as early proto-systems but treated as speculative.

Phoenician Alphabet and Its Influence

• Multiple comments explain that Phoenicians did not invent writing, but developed and spread an alphabet derived from older North-Semitic and Egyptian systems.
• The Phoenician script is described as an abjad (consonant-only), reduced from ~27 to 22 letters as certain sounds merged.
• This 22-letter set was too small for many languages; Greeks added vowels to create the first “full” alphabet, and later Latin followed.
• Many modern alphabets, including some in South Asia, are said to trace back to the Phoenician lineage.

Non-literate Societies and Oral Traditions

• Examples given: Aboriginal Australians, various pre-industrial societies, and some European groups whose languages only recently gained written forms.
• Aboriginal cultures are described as rich in oral transmission and symbolic drawing (e.g., sand drawings, rock art) despite lacking a formal script.
• A 1969 cross-cultural study is cited: ~39% no writing, ~37% pictures only, ~24% writing.

Age and Independent Emergence of Writing

• One side emphasizes strong archaeological evidence that writing is recent (proto-writing ~9,000 years, true writing ~5,000) and tied to complex agricultural states.
• Skeptical voices argue that absence of evidence isn’t proof of absence and note the possibility of lost or undiscovered records.
• Discussion covers Mesopotamian token systems, Egyptian hieroglyphs, Chinese and Mesoamerican scripts, and the role of accounting.

Shipwreck Preservation and Material Decay

• Several comments note that preserved wrecks are exceptional; in most waters, wood is quickly destroyed by organisms and physical forces.
• Good preservation occurs in special conditions: deep ocean, low-oxygen seas (Baltic, Black Sea), or burial under sediment/sand.
• This particular wreck was protected by sand and later a metal “coffin”; moving it starts a race against decay, requiring multi-year conservation with specialized treatments.

Related Museums and Comparative Wrecks

• Commenters recommend visiting shipwreck museums: Steamboat Arabia (USA), Vasa (Sweden), Mary Rose (UK), and a Baltic wreck museum (Sweden).
• Some find the conservation techniques—drying and resin impregnation—as interesting as the ships themselves.

Numeric Precision and “2.6k” vs “2600”

• A side thread debates whether “2.6k” vs “2600” conveys precision or approximation and how significant figures should be indicated.
• Multiple conventions are discussed, and it’s noted that, in context, both are clearly approximate for the age of the wreck.

Export controls & compute threshold

• Central rule: export controls on releasing model weights trained above 10^26 operations; several comments convert this to very large GPU clusters and call it arbitrary or already obsolete.
• Critics argue compute isn’t a stable proxy for danger: algorithmic advances and test-time (inference) compute can make smaller or more efficiently trained models very powerful.
• Supporters see FLOP caps as an imperfect but measurable first step, analogous to controlling high-end night vision or radar; better than waiting for a “perfect” metric.

Effectiveness, circumvention & crypto-war analogies

• Many doubt enforceability: model weights can be exfiltrated via hacking or insiders; cloud KYC and security are seen as only mitigations, not real barriers.
• Historical analogies to 1990s crypto export controls: expectations of workarounds (book-printing of code/weights, steganographic encodings), and risk of pushing innovation offshore.
• Some argue that even slowing adversaries by months and forcing them to spend more on domestic chips is worthwhile; others say this just accelerates import substitution and Chinese GPU ecosystems.

Geopolitics, China & military framing

• Widespread view that the rule’s real purpose is to deny dual‑use AI (e.g., autonomy, targeting, drones) to adversaries.
• Strong disagreement over whether the US still has a meaningful “military advantage,” and whether China is already leading in open‑weight LLMs and efficiency.
• Debate over whether collaboration with China reduces conflict or simply empowers an illiberal superpower; some respondents flip this, viewing the US as the greater global aggressor.

Impact on innovation & open source

• Fear that limiting US open‑weights above the threshold while Chinese labs are unconstrained will hand long‑term open‑source leadership to China.
• Others counter that these rules are explicitly meant to “stifle innovation” abroad, not at home, and mainly apply to frontier-scale training.

Country tiers & alliances

• Framework splits countries into three tiers with differing restrictions.
• Some close allies and EU/NATO members fall into a restricted middle tier, which is perceived as insulting or treating them as “cheap brainpower.”
• Unclear how this meshes with EU single‑market rules or how exceptions/overrides will work.

Other concerns

• Minor thread on federal sites leaking visitor data via Google Analytics and the privacy implications.
• Underlying divide: some assume AI will not become extremely dangerous soon and see the rule as overreach; others assume near‑term, extreme capabilities and think the regulation is timid.

Fisher’s Brilliance vs. Fallibility

• Many emphasize Fisher’s extraordinary impact on modern statistics and genetics, listing a large number of concepts and methods that bear his name.
• Others push back against “great man” narratives, arguing discoveries are often inevitable and over-crediting individuals distorts history.
• A middle view: individuals can be genuinely brilliant and still make severe errors; brilliance raises the average quality of reasoning but doesn’t prevent outliers that are “boneheadedly wrong.”

Why a Top Scientist Can Be Deeply Wrong

• Suggested factors: financial incentives, ideological alignment, identity (e.g. as a smoker or conservative), and the overconfidence that comes from a lifetime of being right.
• One recurring theme: smart people are especially good at rationalizing desired beliefs, including contrarian positions.
• Some argue that once someone is speaking from greed or ideology rather than reasoning, we should stop treating them as “brilliant” in that context.

Debate Over Smoking Evidence and Causality

• Most commenters treat smoking’s harms as overwhelmingly established, and see contrarianism here as misplaced and often exploited by industry.
• A minority voice questions the intensity of anti-smoking consensus and calls for re‑examining raw data and methods, including possible benefits of very light use (e.g. appetite suppression, social function).
• There is a meta‑debate on epistemic standards: whether believing polonium-laden smoke is harmful requires RCT-level proof, or whether established radiation biology suffices.

Genetics, Statistics, and Fisher’s Specific Claims

• Fisher’s core argument is summarized as: a genetic predisposition could both increase smoking and lung cancer risk, creating correlation without smoking being causal.
• It’s noted that such genes do exist, but later work argued their effect size is far too small to explain the observed association.
• Fisher is criticized for clinging to one early finding about inhalation and for underweighting accumulating contrary evidence.

Eugenics and Moral Evaluation

• Thread documents Fisher’s deep early involvement in organized eugenics and argues this justifies calling him a eugenicist.
• Others point to later statements supporting equal rights but opposing racial mixing, debating whether his views evolved and how much that matters.
• Broader context: eugenics and racial separatism were mainstream among many early‑20th‑century elites, but also directly tied to racist policies and worse.

Meta: Expertise, Hubris, and Forum Norms

• Several parallels are drawn to other eminent scientists and technologists who later made confident but dubious claims in unrelated domains (“Nobel disease,” “engineer’s disease”).
• Commenters stress that expertise is domain-limited and that success can breed hubris.
• Significant subthread about HN moderation and tone: how to correct factual errors without personal swipes, and why strict norms exist to avoid flamewars.

US–Dutch Alignment on Export Controls

• Many see the Dutch move as pressure-driven alignment with US policy rather than free choice; others argue it reflects genuine strategic alignment and dependency on US security.
• Several comments emphasize that ASML depends heavily on US components, IP, and markets, so non‑compliance risks sanctions, supply cutoffs, and broader political consequences.
• Some frame it as a concession to an ally; others as a small country being “strong‑armed” with little in return.

Who Owns EUV IP & What Leverage Exists

• One major thread disputes whether the US government “owns” EUV IP.
  • One side cites the 1990s DOE national-lab research and a Cooperative R&D Agreement (CRADA) in which the US retained ownership and licensed the tech via the EUV LLC/Silicon Valley Group, later acquired by ASML.
  • Others counter that: ASML has invested decades and billions into its own implementations; early patents may have expired; much of the system is European-developed; and key details of the licensing contracts are undisclosed.
• There is agreement that export controls and licensing conditions tied to US-origin tech create durable leverage, independent of patent expiry.

Geopolitics, NATO, and the ICC

• Multiple comments stress that the Netherlands is a “port economy” reliant on US‑protected sea lanes and NATO guarantees, including F‑35 purchases and nuclear sharing.
• A heated subthread debates the “American Service-Members’ Protection Act” and whether the theoretical threat to invade the Netherlands over ICC prosecutions is credible or political theater.
• Broader discussion laments transactional US rhetoric on NATO and argues Europe should increase defense spending and possibly develop more autonomy, with some warning this could also raise escalation risks.

ASML’s Technological Role

• ASML’s EUV machines are described as among the most complex industrial systems ever built, integrating extreme-precision optics, plasma light sources, and ultra-fast wafer stages.
• Commenters highlight that ASML is less a purely Dutch miracle than a global integrator of US, European, Japanese, and Taiwanese contributions, kept alive over decades by massive funding from major chipmakers.

China, Taiwan, and Strategic Stakes

• Export controls aim to keep China multiple nodes behind leading-edge chips, limiting military capabilities, especially for AI and autonomous systems.
• Some argue this makes Taiwan’s TSMC an even more critical “jewel,” though others note rumored sabotage/remote-disable plans and the difficulty of operating captured tools without ongoing support.

Modern Git commands & safer workflows

• Many advocate preferring git switch / git restore over git checkout to separate “change branch” from “restore files,” though some note these commands are still labeled “experimental” in docs.
• Several suggest avoiding git reset --hard except when you really understand the consequences, recommending:
  • git branch -f <branch> <commit> or git reset --soft/--mixed to “move” branches without nuking work.
  • git reset --keep to retain uncommitted changes when moving HEAD.
  • git stash/stash --all as a safer “clean tree” approach than reset --hard + clean -xfd.

Mental model, reflog, and low-level tools

• Strong focus on Git as a graph of commits and refs; confusion arises from the CLI mixing low-level “plumbing” and high-level “porcelain.”
• git reset variants are widely described as hard to grasp; the “Reset demystified” chapter is frequently cited (within the thread) as essential reading.
• git reflog is highlighted as the real safety net for “oh shit” moments, letting you recover previous states; some even copy the .git directory as an extra backstop.

Alternatives and UX: jj, GUIs, TUI

• Jujutsu (jj) gets repeated praise as a Git-compatible VCS with simpler, more consistent commands (e.g., jj undo, jj op log, jj split, jj rebase flows).
• Many argue beginners should start with visual tools (Fork, GitHub Desktop, Magit, lazygit, etc.) to see the commit tree and make rebases/cherry-picks more intuitive.
• Others warn GUIs hide concepts, leading to deeper confusion when something breaks and only the CLI or reflog can fix it.

Recipe sites, help culture, and risk

• The “Oh Shit, Git” style of copy-paste recipes is polarizing:
  • Fans: great quick references, especially without in-house Git experts.
  • Critics: generic recipes can subtly corrupt history or lose work; better to ask experienced teammates or rebuild a small personal repo than run commands you don’t understand.

Philosophy of commits, branches, and other VCS

• Debate over local vs remote branches: some advocate “all branches are remote,” frequent pushes, and aggressive reset --hard; others value private WIP branches and curated public history.
• Several emphasize frequent local commits (even messy) and then cleaning up via rebase/squash before sharing.
• Comparisons arise: some say Git is powerful but poorly designed at the UI level; others defend it as “not that hard” if you learn the core concepts. SVN and Mercurial are mentioned as more intuitive but less dominant.

Overall reaction

• Commenters express shock, sadness, and a strong sense of losing a singular creative voice; several compare the feeling to Bowie’s death.
• Many note how personally formative his work was and emphasize gratitude that Twin Peaks: The Return was completed before his health declined.

Notable works & viewing recommendations

• Frequently praised: Mulholland Drive, Lost Highway, Blue Velvet, Eraserhead, The Elephant Man, Wild at Heart, Inland Empire, The Straight Story, Fire Walk With Me (plus “The Missing Pieces”).
• Twin Peaks (especially S1 and S3) is repeatedly called one of the best TV series ever made.
• Some recommend starting with more accessible works (The Straight Story, Elephant Man, Mulholland Drive, early Twin Peaks) before diving into Eraserhead or Inland Empire.

Style, themes, and interpretation

• Consensus that his films emphasize emotional and dreamlike experience over clear, complete explanation.
• Viewers discuss the tension between “there is a decipherable plot” vs. “don’t try to solve everything; focus on feelings.”
• Common readings: Eraserhead as fear of parenthood / loss of “the art life”; Lost Highway as unbearable remorse or bad conscience.
• Many highlight his sound design, music choices, and unforgettable imagery.

Dune and adaptation debates

• Strong split on Lynch’s Dune: some call it their favorite version with superior mysticism and atmosphere; others say it’s a fascinating failure and a poor adaptation.
• Long subthread on adaptation theory: film vs. book pacing, necessary changes, and comparisons to Villeneuve’s Dune and other adaptations (The Shining, Starship Troopers).

Twin Peaks & The Return

• S1 widely praised; S2 seen as uneven but redeemed by its finale and Fire Walk With Me.
• The Return is called both his best work and, by a minority, unnecessary or unsatisfying; many stress its experimental structure and emotional power (especially episode 8).

Other work & public persona

• Discussion of his music, painting, sculpture, commercials, web projects, daily weather and bingo videos, and comedy appearances.
• Many share short clips of interviews, rants, and cooking videos as windows into his humor and character.

Transcendental Meditation & foundation

• Multiple links and anecdotes about the David Lynch Foundation, which funds TM instruction for veterans, students, and others; some see this as his most important work.
• A few express skepticism that readers will “take it seriously,” but most mentions are positive or neutral.

Health, smoking, and unfinished projects

• Thread notes his emphysema, lifelong smoking, and possible aggravation from LA air and wildfires (causal links are speculative).
• Mention of at least two rumored but unfinished projects (a film and a Netflix series), with hope that some material might eventually surface.

Critical views & controversies

• Not all reactions are reverential: some found Blue Velvet or Eraserhead nearly unwatchable or disturbing in a negative way.
• One comment notes his signature on a letter supporting Roman Polanski, flagged as a permanent mark against him.

Tourism, Daily Life, and Culture in Albania

• Multiple commenters praise Albania as beautiful, affordable, and welcoming, with mountains in the north and beaches comparable to Greece/Italy.
• Albanians are widely described as friendly and trusting; low levels of petty crime are emphasized (e.g., unlocked cars, unattended purses).
• The Albanian language (Shqip) is highlighted as a unique Indo-European branch; many Albanians understand Italian (via TV under communism) and younger people often speak English.
• Strong pro‑US sentiment is noted, linked to Woodrow Wilson’s role in preserving Albanian statehood and later US support in the Kosovo war (e.g., Clinton statues, street names).
• Travel tips: Tirana’s BUNK’ART museum, Skanderbeg Square walking tour, the Blloku district, Berat, Gjirokastër, and the coastal drive via Vlore/Himarë are frequently recommended.

Bunkers: Scale, Use, and Military Value

• Commenters confirm bunkers are “everywhere”: in cities, countryside, mountains; mostly small, two‑person domes, with some larger complexes.
• Repurposing includes cafés, tourist attractions, and informal uses (including as romantic hideouts and playgrounds).
• Debate on military usefulness:
  • Critics call isolated bunkers tactically weak, loud enough to deafen occupants, and hard to resupply or integrate into modern warfare.
  • Others argue massed bunkers, often built in lines and linked by trenches and hardwired comms, could slow an invader even when empty.
• Several anecdotes describe extreme tests and paranoia around construction quality, with severe punishments for perceived sabotage.

Hoxha, Dictatorship, and Post‑Communist Nostalgia

• Hoxha’s regime is portrayed as highly paranoid, repressive, and economically destructive (bunkers consuming huge resources; widespread interrogation, camps, and executions).
• Some Albanians reportedly retain nostalgia for the dictatorship, often framed as longing for stability and social equality despite repression. Others say most are glad for democracy and EU prospects.
• The bunker program is interpreted by some as both fear‑driven defense policy and a way to channel national effort into low‑intellect “busy work” that posed no political threat.

Crime, Economy, and Comparative Politics

• One subthread claims Albanian organized crime is globally significant and deeply intertwined with the state; others focus on ordinary Albanians’ hospitality and normalcy.
• Broader discussions compare authoritarian systems (Albania, USSR, China, DPRK, Yugoslavia), emphasizing:
  • How ordinary people adapt and comply under fear.
  • Post‑communist corruption and disillusionment.
  • Tension between stability, freedom, and economic development.

Urban density and rebuilding strategy

• Many argue burned single-family homes should be replaced with denser housing (small apartment buildings, row houses, medium density) to improve affordability and support local businesses.
• Others counter that these are some of the most desirable, high-value coastal/hillside areas; market pressure will still push toward luxury units, even if density increases.
• Some insist these neighborhoods will remain exclusive; large-scale “urbanist” redevelopment (e.g., mixed-income, walkable projects) is seen as politically unrealistic there.

Self‑driving cars, parking, and LA car culture

• One camp believes self-driving services will be widespread within a decade, justifying reduced parking requirements.
• Skeptics argue current services are geographically limited, rely on remote interventions, and are too expensive versus car ownership; they see cultural reliance on cars as deeply entrenched.
• Others note that even if AVs expand, high land values in coastal areas mean unit prices will remain high regardless of parking rules.

Economics of luxury vs. affordable housing

• Repeated point: developers will seek to maximize returns; in $3–80M-home areas, that means luxury SFH or luxury condos, not sub‑$3k rentals.
• Some say multiple units on the same land should be more profitable than one mansion; rebuttals stress topographic limits (hills, canyons) and zoning/approval constraints.
• Several argue that without rent caps or profit constraints, “cheap apartments” on prime land are unrealistic.

Where and how to rebuild

• Some advocate not rebuilding in high-fire zones at all, or at least restricting SFH sprawl there.
• Others emphasize topographic limits: much of the burned area is steep hillside or mesas, poorly suited for large-scale transit‑oriented density.
• Counterexamples from other hilly cities (San Francisco, Italy, Japan) are used to argue that terrain is a political, not technical, barrier.
• Fire‑resistant construction (concrete, metal roofs, ember‑resistant design) is discussed; experiences from other countries suggest extreme fires can still overwhelm such measures.

Governance, zoning, and power

• Debate over whether disasters should be used as leverage to upzone rich fire‑prone areas versus focusing on transit‑served parts of the metro.
• Strong disagreement on whether landowners or “leaders” should decide post‑fire land use, and on how much government coercion underlies zoning and building codes.

Customer Experience & Usability

• Most commenters describe the fridge/freezer screens as actively hostile: blocking direct view of stock, adding delay, and often showing incorrect items or prices.
• Several note that doors frequently appeared “full” on-screen while shelves behind were empty, making the disappointment worse than simply seeing an empty shelf.
• Some users stopped visiting stores that implemented them, citing frustration with slow “wake up” times and having to open multiple doors to find items.
• People compare the experience to airport gate screens that show ads before boarding info: ads obstruct the primary purpose of the display.

Advertising Logic & Sales Impact

• Commenters understand the theoretical rationale: point‑of‑purchase ads, impulse buys, and dynamic pricing, similar to online upsell flows.
• Many think this logic breaks down when the customer is already at the shelf; advertising pizza rolls to someone already staring at the pizza roll door seems redundant.
• Several draw parallels with Amazon/YouTube/etc. continuing to recommend items (e.g., toilets, appliances, vacuums) long after purchase, viewing it as bad targeting that persists because brands overpay for low‑quality eyeballs.

Corporate Governance, Pilots & Incentives

• Many see this as a failure of basic judgment and incentives: executives approve a 10‑year, large‑scale rollout for a solution to a “non‑problem.”
• Some note there was a small pilot and reported ~5% sales lift, but suspect novelty effects or manipulated data, and question whether Walgreens independently validated results.
• The close relationship between the startup and a former Walgreens CEO is viewed as cronyism; commenters doubt the startup would have survived without that connection.
• A number argue this should have stayed a small, cheap experiment, not a $200M commitment.

Technical, Cost & Energy Concerns

• Complaints include high heat output from the screens, implying increased energy use for both displays and refrigeration.
• Reports of devices failing (black/white screens, misalignment with shelves, even occasional fires) reinforce views that the tech was immature and over‑engineered for the task of “being a window.”

Privacy, Surveillance & Ad Creep

• Some speculate about future integrations with personal data (device IDs, data brokers, insurance) for targeted ads, highlighting discomfort with more surveillance in physical spaces.
• The project is framed as part of a broader trend: physical retail adopting the web’s pop‑ups, tracking, and “hostile design,” with the warning that user tolerance for such hostility is finite.

Alternatives & Bigger Picture

• Commenters suggest more user‑centric innovations: better lighting, simple e‑ink shelf tags, or app‑based shopping aids (store maps, running totals, scan‑and‑go).
• There’s broad sentiment that filling the world with more ads and screens, without clear benefit to customers, is inherently a bad direction for retail.

Short-lived certificates & revocation

• Six‑day certs omit OCSP/CRL URLs; commenters note this is explicitly allowed for “short-lived” certs (≤10 days now, ≤7 days from 2026) under CA/Browser Forum rules.
• Revocation is seen as largely ineffective in practice: many clients ignore it, and past incidents (e.g., Heartbleed, leaked keys in support dumps) showed revoked certs still being used without user warnings.
• Shorter lifetimes are framed as a pragmatic way to reduce risk from undetected key compromise, though some see the threat model as narrow and benefits as modest.

Operational impact and automation

• Many argue 90‑day (and shorter) lifetimes are meant to force proper automation; if certs break every few days in staging, bugs get fixed earlier.
• Others report real friction: flaky automation, services that require restarts to load new certs, segmented networks, and offline/edge use cases.
• Hobbyists and small setups sometimes find renewals the least reliable part of their stack, provoking frustration with “opinionated” lifetime choices.

Outage and rate‑limit concerns

• Some fear a major Let’s Encrypt outage could break a large fraction of the web with 6‑day certs.
• Counterpoints: week‑long outages are seen as unlikely; robust clients can fall back to other ACME CAs. Skeptics worry that mass fallback could overload those CAs.
• With 5‑per‑7‑days limits for identical hostname sets, very aggressive renewal schedules (e.g., daily) may hit rate limits unless carefully managed.

IP address certificates & BGP/RPKI

• Attack surface is debated: many note BGP hijack–based issuance attacks already exist for domain certs using HTTP/TLS challenges.
• Proposed mitigations include multi‑vantage validation and tying issuance to IPs whose origin AS participates in RPKI; feasibility is unclear.
• LE’s six‑day max for IP certs is justified as limiting abuse, especially in cloud environments with frequently recycled IPs.

Use cases and limitations of IP certs

• Use cases mentioned: DoH/DDR resolvers, easy-to-remember diagnostic IPs, cloud demos where getting a domain is bureaucratically hard, and bootstrapping OAuth/tunnel tooling without a domain.
• No public CA support for private RFC1918 IPs; suggestion is to use DNS names (possibly internal) or private CAs.

Certificate Transparency & ecosystem tooling

• Short-lived certs will greatly increase CT log volume; CT monitors already face scalability issues but want full historical records.
• Some ACME clients and servers (e.g., Caddy, CertMagic, others) are already adding support for ACME profiles and short‑lived cert workflows.