Hacker News, Distilled

Perceived Goals and Incentives

• Several comments frame the policy as a predictable outcome of democratic incentives: large voting blocs (older people, parents) tend to attract subsidies.
• Others see it as a practical way to unlock more parental labor supply while “keeping the money in the family.”
• Some suspect a political shift: from earlier pushes to equalize parental leave between mothers/fathers to newer incentives that may subtly reverse that trend.

Fraud, Enforcement, and Administration

• Some worry about fraud risk but argue the monetary scale and neediness of potential fraudsters make it tolerable.
• Others are more concerned about heavy-handed anti-fraud regimes, citing past scandals in other European childcare-benefit systems.
• It’s noted that grandparents still need to prove the family relationship, limiting the most blatant abuse.

Role of Grandparents and Family Structures

• Many describe grandparents already doing 1+ days of childcare, often informally and unpaid; some welcome formal compensation, especially given the physical toll.
• Others note big variation: in some cultures grandparents are central caregivers; in more individualistic or affluent contexts they may prioritize travel or their own activities.
• Cases are mentioned where grandparents are primary caregivers because parents are absent or dysfunctional.

Fertility, Demographics, and Pronatalism

• Debate over whether such benefits materially raise fertility:
  • One side claims Scandinavian-style supports have helped slow declines, and that playing with grandkids has side health benefits.
  • Another side points to long-run fertility rates (often ~1.3–1.6), arguing pronatalist policies are very expensive and only modestly effective.
• Cited research links women’s education, autonomy, and contraception access with lower, later fertility and suggests that reversing fertility decline would require major, unlikely systemic changes.
• Some argue fears of “extinction” are overstated globally, but others focus on regional demographic decline (e.g., Europe, Japan).

Economics, Work, and the Two‑Income Ratchet

• Several comments connect childcare policies to encouraging dual‑income households, which can raise living costs and make single‑income family models harder to sustain.
• Others counter that dual incomes are simply people’s preferences and a route to financial independence, not a “plot,” even if societal costs emerge.

Cultural Context: Sweden and Beyond

• Sweden is described as having strong, subsidized childcare and long, flexible parental leave; many see it as an excellent place to be a parent.
• Some characterize Scandinavian societies as highly individualized and “atomized,” with early home-leaving and fewer multigenerational households, making “grandparents-as-a-service” feel like a logical extension.
• The “Swedengate” debate (kids not always being fed at playdates) is cited as an example of cultural norms around family boundaries differing from other countries.

Age of Parenthood and Grandparent Capacity

• Multiple comments note rising parental ages: instead of ~22-year-old parents and ~45-year-old grandparents, it’s now more often ~32 and ~65.
• One view: older grandparents may be less physically capable for intensive childcare, especially amid rising obesity and health issues.
• Counterexamples stress that many people in their 60s and 70s remain active, work, and provide substantial childcare; the “65 = dependent” claim is challenged as too broad.
• Several anecdotes show that where grandparents are older or frail, parents compensate with expensive paid childcare, which not everyone can afford.

Broader Social Reflections

• Some see late parenthood and weak extended-family networks as contributing to loneliness, weaker local communities, and stress on marriages.
• Others emphasize that many adults—especially women—rationally delay or limit childbearing to protect careers, autonomy, and lifestyle, viewing multiple pregnancies in their 20s as too high an opportunity cost.

Protecting Less-Tech-Savvy Relatives

• Many argue the most urgent action is helping parents/elderly family set up 2FA/MFA, ideally in person.
• Hardware keys and backup codes are favored for reducing account takeovers, but people stress also planning for account recovery.
• Some describe using shared TOTP secrets within families as an out-of-band way to verify identity during suspicious contact.
• Several urge explicit conversations with older relatives about scams (phishing, “investment,” “romance,” “grandchild in trouble,” gift-card scams).

Effectiveness and Limits of 2FA/MFA

• Commenters note conventional MFA (SMS/TOTP) does not stop phishing sites that proxy logins and steal tokens in real time.
• Hardware security keys / WebAuthn are cited as highly effective against phishing, with one large company reportedly reducing employee phishing to zero after adopting them.
• Others emphasize that none of this helps against scams where the victim is simply convinced to send money.

Real-World Scam Experiences

• Multiple detailed anecdotes: fake ISP support, IRS threats, “tech support” from overseas, deepfake-style Elon Musk crypto promotions, gift-card and Bitcoin QR scams, “your relative is in jail” calls.
• Banks and some retailers sometimes intervene when they detect likely scams (e.g., elderly customers withdrawing large sums or buying many gift cards), but this is inconsistent.
• One story describes a highly orchestrated, multi-day scam that extracted $25k from an elderly victim despite some bank resistance.

AI and Phishing Evolution

• People expect AI to remove “bad grammar” as a phishing tell; others say now overly polished language can itself seem suspicious.
• Some note AI can be instructed to mimic imperfect language, teenagers, or non-native speakers, making detection harder.
• A few report already seeing AI-like phishing and deepfake-style scam videos.
• Others are surprised AI scams aren’t more widespread yet, suggesting reasons: existing low-tech methods are already profitable, AI stacks are not turnkey for criminals, and economics/ROI may not yet favor large-scale AI deployment.

Biometrics, Device Fingerprinting, and Security Debate

• Concerns that AI-driven phishing will replay captured device profiles and behavior to bypass fraud detection.
• Strong criticism of biometrics and behavioral signatures: they are inferrable, can be spoofed, and unlike passwords cannot be rotated after compromise.
• Some argue current approaches are “least bad” given what banks/processors will pay for; others think we’re relying on identifiers (like SSNs) never suited for authentication.

Email, Browsers, and Platform Responsibility

• Several blame email clients and browsers for hiding full email addresses and URLs, eroding users’ ability to inspect links.
• Corporate “safe link” and tracking systems that replace real URLs with long opaque redirects are criticized for training users to click unreadable links while claiming to improve security.
• Some see this as mainly about marketing/tracking rather than safety, and describe tension inside companies between security and marketing priorities.
• Broader frustration that “legit” businesses increasingly resemble scammers in UX and communication style, shrinking the gap between real and fraudulent messages.

Telecom, Robocalls, and Voice Spoofing

• People ask why carriers don’t block foreign-origin robocalls or caller-ID spoofing more aggressively; one answer is that carriers profit from every call.
• There is growing anxiety about voice cloning: even a single recorded “yes” or short call could be misused for social engineering or voice-based authentication.
• Some recount near-misses where only a Western Union clerk or bank employee stopped a “frantic relative” payment scam; they note this was possible even years ago, before current TTS advances.

Education and Social Response

• Many advocate a cultural shift akin to “talk to your kids about drugs,” but for scams: a continuous, explicit education effort for older and vulnerable people.
• There is pessimism that high-trust social norms will erode as AI makes it harder to distinguish genuine communication from sophisticated fraud.

Extreme food and gaming “quests”

• Commenters recall long-form projects (cross‑country burrito hunt, barbecue parody, visiting every themed chain location) as fascinating for their sheer excess and self‑imposed suffering.
• Similar appeal noted in grind-heavy gaming content (RuneScape “one‑chunk” accounts, hardcore challenge modes, massive Factorio mods): audiences enjoy watching others pour huge effort into marginal gains.

Tax write‑offs and “turning dining into a business”

• One thread explores starting a minimal food/movie/travel review “business” to deduct tickets, meals, and travel.
• Others push back:
  • IRS hobby‑vs‑business rules and “ordinary and necessary” standards limit what’s deductible.
  • Entertainment deductions have been narrowed; critics are an exception only when clearly working.
  • Debate over whether pass‑through losses can offset other income; some say yes with limits, others say only against that business’s revenue.
  • General consensus: pushing this too far risks audit.

Health effects of restaurant food

• Many emphasize that restaurant cooking leans heavily on fat, salt, and sugar; portion size in the US is repeatedly cited as a major driver of weight gain.
• Several share personal experiences of gaining weight while eating out frequently (travel, catered office meals) and losing weight when eating home‑cooked food.
• A cited study links frequent meals “away from home” to higher all‑cause mortality; others question its lack of control for cuisine type, restaurant quality, and confounders (e.g., travel, occupation).
• Some argue occasional restaurant meals are fine; the problem is routine fast food and oversized portions, not rare fine dining.

Home cooking vs restaurant capabilities

• One side claims certain flavors and techniques (high‑heat pizza, wok hei, deep‑fried fast‑food textures, fine‑dining patisserie) are effectively unreachable at home due to equipment, ingredients, and labor.
• Others counter with workarounds: pizza steels, outdoor ovens, blowtorches, air fryers, and careful technique can get “close enough” for most people.
• Separate subthread on recipes: many home recipes (including from newspapers) are “heavy.” People report routinely cutting sugar/fat and still getting good results, and debate weighing vs volumetric measuring.

Life and health as a restaurant critic

• Multiple comments highlight that four rich review meals a week is physically punishing and leads to weight gain.
• Suggestions include: tasting but not finishing dishes, sharing among a group, spreading the job across several critics, or using weight‑loss drugs.
• A reported experience with such medication describes severe side effects (nausea, dehydration, fainting), countering the idea that it’s an easy fix.

Criticism, popularity, and taste

• Discussion revisits harsh reviews of touristy, gimmicky restaurants.
• Some argue a critic should judge food quality regardless of an owner’s public niceness or charitable work.
• Others note that professional critics’ tastes often diverge from what mass audiences enjoy; high‑end restaurant culture is compared to the art world in its tendency toward pretension.

Insider perspectives from the restaurant world

• Restaurateurs describe the stress of realizing a major critic is in the dining room, including tactics to identify critics, showcase more dishes, and fill empty rooms.
• A positive review is said to make reservations scarce for weeks.
• Former industry workers who moved into tech say the experience gave them deep insight into consumer behavior and empathy for users.

Scope and Vision of the New AI+Education Company

• Building an “AI teaching assistant” rather than a full teacher replacement.
• First product: an LLM-focused course (LLM101n) intended to be a very high‑quality, hands‑on intro to building LLMs.
• Longer‑term vision: personalized, interactive courses where an AI guides students through carefully designed materials, with analogies like “learning physics with a virtual Feynman”.
• Some see this as overlapping with or competing against existing efforts (e.g. Khan Academy’s Khanmigo, Synthesis, MOOCs), others think the initial focus is more on motivated adults and AI/ML content.

AI as Tutor vs. Human Teacher

• Many commenters report strong personal success using LLMs as on‑demand tutors for math, physics, programming, and language learning, especially as adults.
• Consensus that AI works best for self‑motivated learners; much less agreement about effectiveness for typical K–12 students.
• Several teachers emphasize that classroom reality is dominated by behavior management, social dynamics, and unstable home environments; AI does not solve these.
• Human teachers are also valued as role models, enforcers, and facilitators of peer learning—roles that are hard to automate.

Reliability, Hallucinations, and Trust

• Multiple anecdotes of LLMs confidently producing wrong answers (basic arithmetic, physics constants, modular arithmetic, legal case citations, fake references).
• Some argue newer models hallucinate less and are often more accurate than average web pages or even some teachers.
• Others stress that novices can’t detect subtle errors, so using LLMs as primary tutors for children is risky.
• A recurring suggestion: “trust but verify,” use multiple sources, and design systems that offload math/lookup tasks to more reliable tools (e.g. calculators, code).

Motivation, Equity, and Systemic Constraints

• Disagreement over whether most kids are naturally self‑motivated or demotivated by current schooling; Montessori and homeschooling are cited as counterexamples to “kids do nothing without supervision.”
• Many note that pandemic “remote school failures” reflected broader social and parental issues, not just technology limits.
• Concern that AI tutors could become a “low‑cost, low‑quality” track for poorer students while wealthier families keep access to small classes and human tutors.
• Others counter that even “Khan Academy + LLM Q&A in local languages” would be a huge upgrade for many globally.

Business and Pedagogical Challenges

• Education is described as highly resistant to disruption: complex stakeholders, misaligned incentives, and long institutional sales cycles.
• Discussion of whether the viable market is schools/companies vs. parents vs. self‑learners; several see parents and autodidacts as the most promising.
• Doubts that LLMs can genuinely emulate historical greats (e.g. Feynman) given limited data and lack of real understanding.
• Worry that systems may optimize for engagement and satisfaction rather than deep learning; reference to research where students “liked” less effective teaching more.
• Ethical concerns: embedded biases, lack of provable correctness, potential for AI “replacing” parental or mentor relationships with emotionally tuned tutoring.

Enthusiasm and Optimism

• Many express excitement based on the founder’s past educational work and want higher‑production, deeper AI/LLM courses.
• Optimism that AI can:
  • Personalize pacing and difficulty.
  • Generate practice problems and explanations on demand.
  • Provide immediate feedback on writing, coding, and math steps.
  • Free human teachers from some grading/admin work so they can focus more on high‑value interactions.

How the problem was discovered

• Many commenters focus on the irony that such a critical flaw was learned informally at a conference.
• This is compared to “hallway conversations,” Slack browsing, and back‑channel chats where crucial information occasionally appears amid mostly noise.
• Some argue this shows the value of practitioner networks and specialized forums/chats for sharing “everyone knows” issues that aren’t well documented.

Vendor responsibility and disclosure

• Strong criticism of Infineon for not proactively notifying affected customers when their rad‑hard MOSFETs failed to meet radiation specs.
• Others push back, noting:
  • It’s unclear whether NASA bought directly from Infineon or via contractors.
  • B2B law often puts defect‑detection burden on the buyer unless contracts say otherwise.
• Several note that rad‑hard parts are typically bought directly from manufacturers with strict lot traceability, so the vendor likely knew who the major space customers were.
• There is debate over whether this was a one‑off blunder vs. a systemic trust problem in hi‑rel components.

Specifications, QA, and legal nuances

• Discussion around “in spec or not”:
  • Specs are often based on limited testing assumptions (sample percentages, separate test conditions).
  • A part can meet individual voltage/temperature/radiation specs yet fail under combined extremes.
• Some point out that if problems are “open” and not promptly reported, warranty recourse may be lost under certain legal regimes.

Radiation environment and shielding

• Multiple posts stress that Jupiter’s environment is far harsher than Mars or Earth orbit; single‑event and total dose effects are severe.
• Simple “add a thin lead sheet” suggestions are rejected:
  • High‑energy particles can penetrate centimeters of material and create secondary radiation showers.
  • Lead is heavy and can worsen radiation effects in some cases.
  • Spacecraft mass and balance constraints make late shielding changes very hard.

Mitigation options and schedule risk

• Replacing the MOSFETs would require opening a sealed electronics vault inside an already‑integrated and tested spacecraft.
• Commenters note this is technically possible but would trigger extensive rework, re‑qualification, and likely schedule slips, possibly missing backup launch windows.
• Others suggest more aggressive system‑level testing and failure‑mitigation planning rather than full hardware replacement, but outcomes remain unclear.

Legal framework and incentives in Ireland

• Law in the Republic of Ireland: using a metal detector to search for archaeological objects is illegal without prior written government consent; penalties can exceed €63k and include prison.
• Many commenters think this level of penalty makes anonymous donation rational and self-identification irrational.
• Others argue the law is designed to protect archaeological sites from damage, not to be “anti-fun,” and note that using detectors for non-archaeological purposes (e.g., cables) is treated differently.
• Debate over whether it’s illegal only to search for archaeological objects vs. accidentally finding them while looking for something else; outcome is described as legally and practically murky.

Archaeologists vs. detectorists

• Several comments stress that the scientific value lies in context: precise location, surrounding soil, stratigraphy, and associated items.
• Taking artefacts without documentation is compared to scraping paint off a painting and sending flakes; the “site” is what matters.
• Hobbyists counter that:
  • 99.9% of finds are junk.
  • Archaeologists often will never dig those fields anyway.
  • Strict bans alienate detectorists and push activity underground.
• Some report successful collaboration models (e.g., Denmark, UK/Ireland building sites) where detectorists are trained and supervised.

Landowner and practical concerns

• Rural anecdotes: farmers and homeowners sometimes quietly destroy or ignore finds to avoid delays, loss of control over land, or construction hold-ups.
• Commenters argue that if discovering artefacts can freeze development with little compensation, people will rationally hide or destroy evidence.

Trust, prosecution, and anonymity

• Skepticism that museum promises of confidentiality can override legal duties to report crimes; concern about mandated reporting and lack of prosecutorial discretion in some jurisdictions.
• Some suggest only a formal legal waiver or symbolic minimal fine would make it safe for the finder to come forward.

Proposed alternative systems

• Suggested reforms include:
  • Licensing plus mandatory training.
  • Finder’s fees/bounties tied to not disturbing sites further.
  • Clear compensation for builders and landowners.
  • Structured cooperation with detectorist associations.

How Outlook Email Reactions Work

• Outlook/Exchange lets users “react” to emails; for other Outlook users, this shows as inline emoji on the original message.
• For many non‑Outlook recipients, each reaction arrives as a separate “X reacted to your message” email, often with an image that may be blocked.
• One commenter tested and found no simple distinguishing header; reaction emails look like normal mail plus the usual x-ms-* headers.
• Gmail has added a similar emoji reaction feature, so this is no longer unique to Microsoft.

Postfix Header Opt‑Out and Technical Concerns

• The blog’s approach is to add a custom header (e.g., X-MS-Exchange-Organization-DisableReaction: true) in outbound mail so Microsoft servers suppress reactions.
• Some note this “fix” breaks DKIM because the MTA modifies headers after signing.
• Alternatives discussed:
  • Rejecting such mails at SMTP time (sends a server‑generated bounce).
  • Blackholing them after acceptance (nicer for the recipient, confusing for the sender).
  • Simple content-based filtering on reaction subjects/text.
• Several argue Microsoft should have shipped this as opt‑in based on known‑capable clients, not opt‑out via magic headers.

UX, Attention, and Email Culture

• Many participants dislike reactions in email specifically:
  • Extra low‑value messages clutter inboxes and notifications.
  • Ambiguity: a thumbs‑up may mean “acknowledged,” “agree,” or just “saw this.”
  • Email is treated as “serious” async communication where explicit, textual replies are preferred.
• Others like reactions as lightweight acknowledgments, especially inside a single Outlook/Exchange environment, reducing “OK/Thanks” reply‑all spam.
• Some prefer read receipts; others find them invasive or misleading and routinely disable them.

Reactions in Chat, SMS, and Other Systems

• Reactions are generally viewed as useful in Slack/Teams/IM: fast acks, less typing, reduced message noise if clients aggregate them.
• Complaints arise when these systems leak into incompatible channels:
  • iMessage reactions forwarded as SMS text (“X liked ‘…’”) are widely seen as spammy.
  • RCS and Android/iOS now try to parse and collapse such texts, with mixed success.
• Several stress that norms differ by medium: what’s fine in real‑time chat can be confusing or hostile in email.

Microsoft, Standards, and Historical Parallels

• Many see this as another case of “extend” in “embrace‑extend‑extinguish”: leveraging a standard (email) while degrading interoperability.
• Parallels drawn to:
  • Microsoft Comic Chat mangling IRC with extra metadata.
  • Exchange “recall” messages that only worked inside Exchange.
  • Wi‑Fi Sense password sharing and SSID _optout conventions.
• There is mention of an existing RFC for email reactions, but uptake appears minimal; Outlook’s implementation is proprietary.

Ad attribution in “Private” mode

• Many are alarmed that Web AdAttributionKit is now active in Private Browsing.
• Critics argue: private mode should “do nothing for advertisers,” not spend user resources on ad measurement.
• Some see this as capitulation to adtech; others argue it’s a constrained, “privacy-preserving” alternative to worse tracking, and that sites need some ad analytics to survive.
• Debate over whether this is “worse for privacy”:
  • One side: before, private mode had no ad attribution at all; now it does.
  • Other side: it replaces more invasive techniques and doesn’t track individuals.

Tracking, fingerprinting, and GDPR

• Discussion of fingerprinting: only 20–30 bits of entropy plus IP can uniquely identify users; even “cookieless” systems can track.
• Some note this is likely not GDPR-compliant, but enforcement is seen as weak.
• New Safari protections claim to blunt behavior-based signals (typing speed, cursor movement) and add canvas noise; concern this may break games or web photo editing.

Proxying HTTP and DNS; who to trust

• Safari’s multi-hop proxy for unencrypted HTTP and DNS splits knowledge between Apple and a CDN provider.
• Some trust this more than ISPs; others explicitly trust their ISP more than Apple/US companies.
• Comparisons to Tor: more hops and jurisdiction diversity, but still vulnerable to global traffic analysis.
• Disagreement about whether this model increases practical deanonymization risk vs. plain HTTP through an ISP.

Private Browsing semantics

• Several note original “private mode” was about hiding history from local device users, not network/website tracking.
• Expectations have shifted; now people assume strong online privacy, leading to confusion and lawsuits.

Extension breakage and site compatibility

• Advanced tracking protection can break Safari extensions that rely on full URLs.
• Some expect these protections will also break sites, but see that as an acceptable tradeoff.

Comparisons and alternatives

• Firefox lauded for built-in first-party data whitelisting and cookie auto-delete patterns.
• Brave and Firefox’s own ad-measurement features mentioned; Firefox enables a “privacy-preserving ad measurement” option by default but lets users disable it.

General sentiment

• Mixed: appreciation for Apple’s privacy work overall, but strong discomfort that even Private Browsing now includes any ad attribution.

Overall Reaction to the Stunt

• Many find the campaign wildly creative, ambitious, and well-executed, especially for breaking into competitive creative/advertising roles.
• Others see it as overkill for only landing a short contract, questioning whether it was truly a “success” given time, money, and possible reputational risks.
• Several compare it to “red teaming” or hacking a security firm to get hired: using the same tools and tactics as the industry they want to join.

Ethics, Creepiness, and Privacy

• Strong pushback on multiple elements:
  • Use of AI deepfakes of targets’ faces and voices.
  • Hyper-targeted ads via lists padded with deceased people.
  • Physical deliveries in fake uniforms to offices.
• Some describe it as stalking, invasive, or “sociopathic,” and say they’d involve law enforcement if they received such a package.
• Others argue it’s legally permissible (public data, parody-like deepfakes, targeted ads are standard) and less harmful than mainstream adtech practices already in use.
• There’s disagreement over whether this normalizes or selects for increasingly creepy behavior in marketing.

Refugee / Financial Context

• The authors’ framing as displaced people “with only a backpack” is challenged; commentators point out the evident budget (robot dogs, iPads, custom gear) and call the framing misleading.
• Some are sympathetic given the war background and relocation stress; others argue that doesn’t justify boundary-crossing tactics or lavish spending.

Security and Workplace Norms

• Multiple comments note security concerns: unknown iPads, easy access to offices with high-vis vests, and how this could be repurposed for phishing, radicalization, or worse.
• Several say this clashes with UK professional norms and could easily have escalated to police, evacuations, or legal issues.

Reflections on Marketing, AI, and Power

• Thread connects the stunt to broader discomfort with advertising, surveillance capitalism, AI-driven manipulation, and exploitation of personal data (including the dead).
• Some see it as a sharp demonstration of how easily egos can be targeted and how close modern marketing is to social engineering.
• Others, especially from a creative-industry perspective, say this kind of boundary-pushing is exactly what top-tier agencies quietly reward.

Site & Infrastructure Issues

• Original site was “hugged to death” and often unavailable; several people used the Internet Archive mirror.
• One commenter asked about suitable free CDNs for small hobby sites; Cloudflare was the only one with first‑hand experience.

---

Value of the Elizabethan Drama Project

• Strong appreciation for freely available, well‑annotated editions of non‑Shakespeare Elizabethan and Jacobean plays.
• Some see it as the kind of passion project that “renews faith in the internet,” especially for giving cultural, historical, and classical-allusion context.

---

How Hard Is Shakespeare/Early Modern English?

• Experiences diverge sharply:
  • Some find Shakespeare “perfectly comprehensible” modern English once you adjust.
  • Others, including strong readers, find even basic comprehension extremely difficult, describing it as “word salad.”
• Non‑native but highly literate English speakers often report less difficulty, possibly due to language‑learning habits and etymological awareness.

---

Translation vs. Original Language

• One camp argues plays should be translated into contemporary English, comparing this to Beowulf or Dante: otherwise many readers effectively have no access.
• Opponents insist Shakespeare is (early) modern English and that the value lies in the original wording, puns, and verse; translations are seen as flat and “penitential.”
• Middle position: keep originals but provide side‑by‑side modern paraphrases and annotations; this is praised as very helpful.

---

Performance vs. Reading

• Many argue Shakespeare “comes alive” in performance; watching good productions is recommended over reading scripts, especially for beginners.
• Others say even live productions can be opaque without prior reading or notes, especially in dense comedies.
• Several describe specific teaching methods: reading aloud in class, multiple readers, acting, pausing to explain vocabulary and cultural context.

---

Culture, Context, and Allusion

• Several stress that the main barrier is cultural, not purely linguistic:
  • Heavy use of Ovid, mythology, politics, contemporary jokes, and sexual innuendo.
  • Modern students often lack this background, so detailed notes are crucial.
• Some liken reading Shakespeare to “foreign language on easy mode” and find joy in the slightly alien grammar and semantics.

---

Pronunciation, Accents, and OP

• “Original Pronunciation” (OP) is highlighted as revealing lost rhymes, puns, and bawdiness; some find it transformative, others see it as an interesting but niche curiosity.
• Multiple accents (regional British, American, etc.) are seen as valid and often delightful for performance; RP is viewed as just one tradition, historically over‑privileged.

---

Education, Canon, and Alternatives

• Multiple commenters recount being turned off Shakespeare in school by unsupported silent reading and over‑analysis.
• There is debate over whether forcing difficult originals on all students is elitist gatekeeping or valuable intellectual training.
• Some suggest starting with more accessible plays, modern adaptations, or even other early modern authors instead of defaulting to Shakespeare.

Local LLMs and Tooling

• Many recommend Ollama as the easiest way to run models locally; pairing it with Open WebUI (via Docker) gives a friendly browser UI.
• Others prefer more feature-rich UIs like text-generation-webui.
• Alternative entry points: llamafile (single-binary), GPT4All, and direct use of llama.cpp, Exllama, vLLM, or TensorRT-LLM depending on hardware.
• Hugging Face’s open-llm-leaderboard and community tools like the “gpu_poor” site are cited for model rankings and hardware sizing.

Model Sizes, Hardware, and Quality

• 7B models run on modest hardware and are seen by some as “very bad” beyond simple tasks, but others argue they’re remarkably capable for summarization and everyday help given their size.
• 24GB GPUs can run Llama 3 70B in quantized form, though speed and quality claims conflict. Gemma 2 27B is suggested as a strong fit for 24GB VRAM.
• Apple Silicon’s unified memory makes 7B models feasible but slower than dedicated GPUs.

Open-Source LLM Ecosystem (High-Level History)

• Thread recaps a short history from early GPT‑2 era to LLaMA, LLaMA 2, Mistral, Mixtral, Llama 3, and Gemma 2, with quantization and CPU/GPU support (llama.cpp, bitsandbytes, Exllama, vLLM, TensorRT‑LLM) driving local adoption.
• Wrappers like GPT4All and Ollama significantly lowered the barrier to entry.

Codestral Mamba, Mamba Architecture, and Benchmarks

• Excitement centers on a high-profile Mamba2 code model competing with Transformers while offering linear-time inference and 256k-token context.
• Some note that DeepSeek models match or beat Codestral Mamba on several benchmarks and that one table mis-highlights results; CodeGeeX4 is said to surpass them “on paper” but isn’t included.
• Links to primers and explainers on Mamba/state-space models are shared; non-experts find good video and text resources.

IDE and Editor Integration

• For VS Code/IntelliJ, Continue.dev and Sourcegraph Cody are popular; they can use Ollama or cloud APIs, but Mamba2 isn’t yet supported in llama.cpp, so Codestral Mamba isn’t available via Ollama.
• Other options: codegpt.co plugins, TabbyML (with older Codestral), and custom editor scripts (e.g., Vim FIM completion via Ollama).

Closed vs Open Code Assistants & UX

• Open coding models mentioned: CodeLlama, Codestral, DeepSeek-Coder V2, CodeGemma, CodeQwen, WizardCoder, CodeGeeX4; consensus is they still lag GitHub Copilot–class services overall, though some local setups work well.
• Users report mixed but often strong experiences with Claude 3.5 Sonnet for coding and project-scale help; many feel it outperforms GPT‑4o in practice despite benchmarks.
• Several dislike Copilot’s perceived decline in quality and explore alternatives like Supermaven, but pricing and token-based limits cause confusion and frustration.

Context Windows and Long-Context Behavior

• Mamba’s 256k tested context is praised, though some question why it’s lower than Gemini’s claimed million-token range.
• Participants discuss that newer models handle long context better than older “lost in the middle” behavior, but best practice remains to keep key instructions at the beginning or end.

Miscellaneous

• Some criticize the product page’s Cleopatra/mamba joke as historically inaccurate and in poor taste.
• Others think Mistral is missing a revenue opportunity by not shipping an official one-click VS Code extension with a clear paid offering.

Charger numbers & reliability

• UK reportedly has ~930k chargers, but only ~65k are public; speeds and uptime aren’t broken down.
• Several commenters argue raw “installed” counts are misleading because many public chargers are broken for months.
• Subsidies often reward installation, not operation; once uptime is tied to subsidies, maintenance improves.
• Concern that some firms are essentially “in the business of installing chargers” rather than running them well.

Home vs public charging behavior

• Consensus: if you have a driveway/garage, EV ownership is easy; you mostly charge at home overnight.
• Many real-world examples of using slow/destination charging at work, hotels, supermarkets, and neighborhood posts.
• Level 1 (120V) is adequate for many US drivers; Level 2 or 230V sockets in Europe/UK give full overnight charges.
• Detached/old garages or long cable runs can make home upgrades costly, but incentives can offset.

Gas stations & oil companies

• Debate why there aren’t more DC fast (L3) chargers at gas stations.
• Arguments: people don’t want to “hang out” at stations for 20–40 minutes; better to charge where they shop or work.
• Practical issues: grid capacity, trenching costs, loss of parking/turnover for convenience stores.
• Some countries (e.g., Germany, UK examples) are already mandating or rolling out high‑power chargers at stations.

Economics & incentives

• Home charging can be dramatically cheaper per mile than petrol; public fast charging can be 4–5× home rates.
• High electricity tariffs in some regions (e.g., parts of California, UK public chargers) erode the cost advantage.
• For low‑mileage drivers, depreciation, higher insurance, and charger install costs can make ICE or hybrids cheaper.

Adoption patterns UK vs Europe

• Some see the UK as lagging “Europe” despite good fundamentals; others note similar new‑sales shares to Germany.
• Proposed reasons: organized anti‑green politics, high share of on‑street parking, low average annual mileage, high power prices, and lingering fears about battery replacement.
• Data cited that ~65% of GB households have or could have off‑street parking, but only a small fraction have chargers.

User experiences: positive & negative

• Positive: several report painless multi‑summer EV road trips across UK/Western Europe, helped by dense DC networks (Ionity, Fastned) and multi‑network cards.
• London and some affluent UK suburbs are described as having reached a “tipping point” with abundant chargers (streetlights, 22kW posts, 250kW hubs).
• Negative: one UK household running a Polestar 2 without a home charger found it a “nightmare” due to slow home‑socket charging, expensive/broken public chargers, and winter range; returned to ICE.
• Non‑Tesla long‑distance travel is often portrayed as stressful: app juggling, broken chargers, and backup plans, versus Tesla’s integrated routing.

Battery longevity & charging speeds

• Multiple anecdotes of high‑mileage Teslas with modest degradation even with heavy supercharger use; fast charging is seen as less harmful than many fear.
• Some still speculate about long‑term costs for used EVs, especially older models with weaker thermal management.

Payment, access & anonymity

• Strong dislike for proprietary apps and “secret club” networks; calls for universal contactless card payment on all chargers.
• Complaints that parking/payment systems are fragmented and often more expensive via phone.
• On anonymity: commenters argue you’re already tracked via pervasive ANPR/CCTV at petrol stations; only basic outlets offer any real anonymity.

Embodiment, “Languaging,” and the Paper’s Core Claim

• Paper argues LLMs lack embodiment, interaction with the real world, and “linguistic agency” (multiple simultaneous goals in communication).
• Some see this as basically correct but obvious and somewhat tautological: it restates what skeptics already believe.
• Others say it’s dated: ignores multimodal models and extensive interactive post‑training, so the video‑game / “brain in a vat” analogy is too rigid.

Training, Feedback, and Post‑Training Methods

• Discussion of RLHF, RLAIF, and newer methods like DPO; all rely on human preference data but differ in how reward is modeled.
• User–model conversations likely feed future training; this blurs the paper’s framing of LLMs as trained only on static corpora.
• Some note that long chains of interactive correction and retraining don’t yet have a standard name; it’s just “how training works.”

Capabilities vs. Limitations

• Several commenters report long, coherent dialogues with recent models, contradicting the paper’s example where the model “loses the thread” quickly.
• LLMs can often perform abstract reasoning on novel, symbolic problems, especially in idealized textbook forms.
• Critics counter that failures at simple arithmetic and brittle reasoning show a lack of underlying concepts; successes are attributed to pattern matching on recurring forms.
• There’s dispute over whether next‑token prediction inherently precludes internal world models; some argue any computable process can be cast as such, others insist current systems are just high‑dimensional curve fits.

Intelligence, AGI, and Definitions

• Repeated theme: we lack precise, agreed definitions of intelligence, consciousness, and AGI, making “LLMs can’t be AGI” or “LLMs think” claims hard to settle.
• One camp: sufficiently advanced behavioral mimicry just is the thing (language, intelligence) under physicalism.
• Other camp: embodiment, stakes, and non‑linguistic experience are essential; text‑only models can at best approximate.
• Some suggest using consensus and obviousness (as with recognizing “flight”) as a pragmatic criterion for intelligence; others point out historical failures to recognize the intelligence of animals or other human groups.

Hype, Value, and Research Trajectory

• Practitioners describe concrete but narrow wins: using LLMs for text structuring and data engineering vs. over‑engineered “agents” without clear business needs.
• Disagreement over scaling laws: some think more data/parameters will eventually hit hard limits; others expect continued gains with better training and hybrid architectures.
• Overall, thread balances excitement about practical capabilities with skepticism about strong claims of understanding, agency, or inevitable AGI.

Scope of “Worthwhile” and Self-Pressure

• Many argue the premise is flawed: “worthwhile” is subjective and adding that filter too early kills ideas.
• Several say you don’t need to code outside work; your job can just be a means to live and fund other passions.
• Hustle-culture expectations (turning job into identity/hobby) are criticized; competence at work is enough.

Programming as Personal Craft vs Public Impact

• Some enjoy writing small, purely personal tools (scripts, home proxies, custom apps) with no audience or business goal.
• Suggested approach: solve tiny annoyances (“I wish it was easier to…”) rather than chasing big/novel problems.
• Others recommend contributing to open source, documentation, tutorials, or developer tooling instead of starting from scratch.

Look Beyond Software for Meaning and Ideas

• Strong theme: get hobbies unrelated to programming—sports, music, woodworking, gardening, tabletop games, travel, creative writing, art.
• Physical or hands-on activities (building furniture, electronics, Arduino, embedded systems) are described as surprisingly satisfying.
• Non-tech pursuits often later generate authentic software ideas (e.g., dungeon planners, music tools, home automation).

Serving Others and Community Engagement

• Recurrent advice: help people directly—volunteer, support local organizations, schools, scouts, political groups, housing efforts.
• Many suggest asking people around you what they struggle with and using tech to assist them.
• Climate tech and other real-world problem domains are highlighted as rich in meaningful, software-adjacent work.

Creativity, Boredom, and Mental Health

• Multiple comments say ideas come from rich inputs: reading widely, new experiences, movement (walks, hikes), and “doing nothing” without screens.
• Boredom is framed as a gateway to creativity if distractions are reduced.
• One thread suggests considering mild depression when feeling purposeless; others push back against casual “diagnosis” but agree reflection and care are important.

Practical Idea-Generation Tactics

• Keep a notebook of observations and reactions to build a “memory bank” for future projects.
• Start with small experiments and iterate; don’t wait for a perfect idea.
• Techniques mentioned: remove a step from common workflows, be your own impatient customer, partner with more “idea-heavy” people, or use “steal my ideas” lists and structured career-guidance resources.

Role of Rust vs Other Languages

• Many see Rust as a major step up from C/C++ for safety while retaining low-level control, especially for security-critical / systems work.
• Others argue it’s only a big improvement for some domains; embedded and interop-heavy work still often favor C/C++.
• GC languages (OCaml, .NET, Java, Go, Elixir/BEAM) are viewed as fine for most apps, but unsuitable where predictable latency, small runtimes, or single-GC constraints matter.
• Several commenters lament that ML-style safety could have been mainstream earlier (e.g., OCaml) if not for GC and unfamiliar syntax.

Interop, Ecosystems, and “NIH”

• Some claim Rust is harder to reuse C libraries in without dropping into unsafe, which they see as negating Rust’s benefits and encouraging “rewrite in Rust.”
• Others counter that Rust was designed for C FFI, that most Rust code already relies on small unsafe islands (including stdlib), and that mature safe wrappers exist for many C libraries.

Async, Executors, and Runtime Design

• Async Rust is a frequent pain point: multiple incompatible executors (Tokio, etc.), difficulty composing async and sync APIs, and “async infecting everything.”
• Some suggest adding a minimal executor or pollster-like tool to std; others point to ongoing work (e.g., keyword generics) to unify sync/async patterns.
• Comparisons are made to GC/runtime fragmentation and to other ecosystems where both sync and async APIs are common.

Syntax, Complexity, Readability

• Complaints focus on sigil-heavy, dense syntax (lifetimes, generics, ::, closures, “turbofish”) and complex function signatures.
• Supporters argue this trades aesthetics for explicitness, C++ familiarity, and “grep-ability,” and that difficulty is more about concepts than symbols.
• Some see a “smaller, simpler, safer language” hiding inside Rust (often “Rust without async”); others point to alternative designs (Yao, Austral, Zig).

Compile Times and Developer Experience

• Slow compilation is widely acknowledged; large projects, macros, and monomorphization are blamed.
• Some prioritize fast edit-compile-run loops and consider Rust’s latency unacceptable compared to C, Zig, or Lisps with fast compilers/JITs.
• Others accept slower builds in exchange for stronger static checking and rely on cargo check and incremental builds.

Adoption, Monoculture, and Ecosystem

• Several argue Rust breaks an old C/C++ “monoculture” by inspiring new systems languages (Zig, Odin, Vale, etc.), making a Rust monoculture unlikely.
• Others see parallels to C++: growing complexity, long compile times, async/runtime issues, and predict C and C++ will remain widely used.
• Some worry about parts of the Rust community being defensive about bootstrapping, specs, or alternative compilers.

Alleged illegal NDAs and SEC issues

• Several comments focus on whether OpenAI’s employee and departure agreements violated SEC whistleblower rules.
• Clauses like “no disclosure unless required by law” are criticized as chilling voluntary reporting to regulators, which SEC has previously treated as a violation.
• The whistleblower letter (linked) alleges waivers of whistleblower compensation and requirements for company consent before contacting authorities.
• Some see this as part of a broader pattern of “move fast, skirt the law” in tech; others note regulators must actually enforce penalties for deterrence.

Use and ambiguity of “safety”

• Multiple commenters say the article’s headline promises concrete “safety risks” but delivers mostly securities/NDAs issues instead.
• The term “AI safety” is described as overloaded and vague: does it mean physical harm, regulatory compliance, financial risk, or Skynet-style catastrophe?
• Some view the safety framing as a PR tool to make systems sound more powerful or to justify secrecy and regulation that entrench incumbents.

AI safety vs moats and corporate control

• A recurring theme is suspicion that calls to “regulate us” are used to raise barriers to entry, harming small startups while leaving big cloud players untouched.
• Others argue secrecy in the name of safety actually worsens safety by hiding real-world abuses (e.g., surveillance, political repression) until after the fact.

Open-source, home-run AI and incentives

• Some commenters say centralized AI should “die” and be replaced by open-source, locally run models for privacy, control, and protection from state or corporate abuse.
• Others counter that most users will always choose convenience and cost over ideals, as with cloud vs self-hosting.
• There is concern about who pays for large open models once hype fades, and whether more efficient, non–brute-force training will be forced by economics.

AGI, extinction risk, and real-world harms

• Strong skepticism that current LLM scaling leads to extinction-level AGI; some call “extinction risk” talk cultish or pure marketing.
• Others note many practitioners do take long-term risks seriously and mention rumored self-improvement/RL work.
• Many argue near-term harms are more concrete: bias and discrimination in automated decisions, AI-powered propaganda and phishing, and potential use by authorities to track or suppress dissent.

Cookie banners, hypocrisy, and GDPR backdrop

• Many comments note the irony that the article’s site (heise.de) itself uses aggressive “pay-or-accept-cookies” banners, possibly in breach of GDPR guidance.
• Users compare different cookie UIs on the site, report dark patterns (single “accept” button, mandatory consent for “free” use), and cite EU/EDPB language that such consent is not “freely given”.
• Broader frustration that GDPR is under‑enforced, leading to nagging banners instead of real privacy.

What Firefox changed and how to turn it off

• Firefox added “Privacy‑Preserving Attribution” (PPA), enabled by default, to measure ad conversions via aggregation/MPC instead of per‑user tracking.
• Desktop: setting is under Privacy & Security → “Web Site Advertising Preferences” → “Allow websites to perform privacy‑preserving ad measurement”.
• Underlying pref: dom.private-attribution.submission.enabled = false (can be enforced via policies or user.js).
• Mobile Android: about:config is hidden; workaround via chrome://geckoview/content/config.xhtml to enable general.aboutConfig.enable, then disable the same pref. Some report variants (Fennec/Mull, F‑Droid builds) where it’s already off or about:config is accessible.
• One Reddit comment (relayed here) claims that if Firefox telemetry is disabled, PPA is also disabled internally, even though the UI still shows it as on.

Consent, trust, and Mozilla’s positioning

• Core criticism: feature shipped opt‑out, with little or no in‑browser notice; users feel this violates Firefox’s privacy‑first brand and ignores existing “strict” tracking/DNT preferences.
• Several see the CTO’s Reddit explanation as “we should have communicated more” rather than admitting the opt‑out default was wrong.
• Some argue Mozilla is pragmatically trying to head off worse ad‑tech/DRM (e.g., WEI‑like lock‑downs) by offering a “less bad” standard; others say advertisers will just use this and continue invasive tracking anyway.
• Repeated pattern noted: Mozilla ships controversial data/ads features quietly (Cliqz, experiments, Pocket, VPN promos), apologizes, then continues similar moves, eroding trust.

Alternatives and forks

• Many propose switching to Firefox derivatives (LibreWolf, Fennec, Mull, Trisquel’s Abrowser), other niche browsers (Orion, Vivaldi, Brave), or future engines (Ladybird, Servo).
• Some prefer to harden Firefox (policies, user.js, uBlock Origin, NoScript) rather than abandon Gecko to a Chromium monoculture.

Funding, governance, and “who Mozilla serves”

• Thread revisits Mozilla’s dependence on Google search money, CEO pay, and failed revenue experiments.
• Mixed views on whether donations or paid Firefox builds could sustainably replace ad‑linked funding, and whether Mozilla’s broader activism distracts from its browser mission.

Additional context and resources

• Commenters link to background on Peter Buxtun and the Tuskegee study, plus podcast series (“You’re Wrong About” and a German podcast “Pandemia”) that give deeper historical context.
• Some note it took years from Buxtun’s first internal complaints to external action.

Moral self-justification and “ends justify means”

• Recurrent theme: people doing harm rarely see themselves as villains; they rationalize with greater-good narratives.
• Several connect Tuskegee to the adage that “the road to hell is paved with good intentions” and to psychological ego-protection against acknowledging wrongdoing.

Was Tuskegee fundamentally racist?

• One side: Tuskegee is cited as a clear case of systemic racism—targeting poor Black men, withholding effective treatment (penicillin) for decades, deceptive consent, and explicit racist statements by officials (e.g., assumptions about “low intelligence” and promiscuity).
• Other side: some argue unethical human experiments also targeted whites (e.g., MKUltra, Operation Sea-Spray) and that researchers may have been driven more by scientific goals than conscious racial malice.
• This framing is challenged as minimizing racism; critics stress that similar atrocities elsewhere doesn’t make Tuskegee non-racist and that stated beliefs and design choices reflected racial hierarchy.

Definitions of racism, race, and group differences

• Multiple attempts to define racism:
  • Treating otherwise-identical individuals differently because of race.
  • Applying group statistics to individuals.
• Debate over whether acknowledging statistical group differences is racist if individuals are treated fairly.
• Distinction drawn between using race as a social/classification variable vs. endorsing “scientific racism” and “race realism.”

US racism and caste analogies

• Non‑US readers describe difficulty fully grasping US racial dynamics.
• Some argue US race relations resemble a caste system (rigid hierarchy, deep historical roots).
• Others say this stretches “caste” too far, noting high interracial marriage rates and emphasizing socioeconomic class instead; opponents respond that historical and ongoing racial stratification still fits a weakened caste pattern.

Politics, power, and moral equivalence

• One thread generalizes from Tuskegee to politics: both major US sides, when in power, believe their coercion is justified.
• Pushback: not all political groups are “equally bad”; equating them can enable worse actors.
• Discussion of “everyone does it” as a self-interested rationalization rather than a genuine moral argument.

Medical distrust and vaccines

• Tuskegee is linked to contemporary distrust of government health programs among African Americans, including lower COVID-19 vaccination rates.
• Another example of abuse of vaccination programs (CIA fake hepatitis campaign in Pakistan) is cited as further eroding trust.

Comparative experiments: Oslo study and Unit 731

• Historical comparison to an earlier Norwegian syphilis study where treatment was deliberately withheld; details and ethics are debated, including whether effective treatments existed at the time.
• Unit 731 is raised to question why Tuskegee continued despite access to brutal Japanese data; reply notes differences in populations, duration, and that Tuskegee was observational (no deliberate infection), while still unethical.

Legal and media tangents (McDonald’s coffee, regulation)

• Side discussion on how US common law and lawsuits substitute for detailed regulation, using the hot-coffee case as an example of media-driven public misunderstanding.
• Debate over whether recent court decisions further weaken regulatory agencies.

Motivations for running models locally

• Privacy and censorship resistance are recurring reasons: users want to run on sensitive data (journals, private audio, “spicy” images) without sending it to large providers.
• Customization is easier locally (changing system prompts, using uncensored models, LoRAs, domain-specific setups).
• Offline and reliable access is valued, especially where connectivity is unreliable or providers could change policy or shut down.

Arguments for cloud-hosted models

• Many note a large quality gap: small local models (e.g., 7–8B) are seen as far behind GPT‑4/Claude-level systems for complex or high-stakes work.
• For productivity, $20–100/month in API usage is argued to be cheaper than buying and operating powerful local hardware, especially once you factor in setup and maintenance.
• Hosted solutions offer integrations (web search, tools like Wolfram Alpha) that local models typically lack.

Cost and hardware trade-offs

• One side: spare hardware + free open models = $0 experimentation; good for students and hobbyists. Cloud is “not free” and can get expensive for heavy use.
• Other side: upfront cost of capable GPUs, electricity, and time is high; for “just messing around,” cheap APIs and free tiers (TogetherAI, Groq, OpenRouter) are seen as better.
• Some argue that for sustained >8h/day workloads, owned or colo hardware can beat big-cloud pricing; others counter that cloud still benefits from economies of scale.

How Exo works and technical feasibility

• Exo uses pipeline parallelism: different devices hold different layers; only activations (embeddings) are sent between them.
• Reported activation sizes: ~8–10 KB per token for 8B models, ~32 KB for 70B; expected to stay O(10–100 KB) even for much larger models.
• On a local network, bandwidth is seen as fine; latency is the main bottleneck, especially over the internet, limiting SETI@home-style global clustering.
• Some users report no speedup when using two MacBooks versus one, suggesting current implementation or scheduling needs work.

Maturity, platform support, and concerns

• Project is explicitly experimental and rapidly changing; issues include:
  • Early hard dependency on Apple-only MLX, conflicting with “everyday devices” marketing.
  • Tinygrad backend exists; llama.cpp support is planned.
  • Windows, iOS, Android, Raspberry Pi, and Coral TPU support are desired but not all are proven.
  • Lack of benchmarks (tokens/sec, latency) and initial missing license; both requested by users.
  • Security model currently assumes a trusted local network; documentation is being updated.

Broader themes

• Debate over whether “swarm compute” of idle devices is desirable versus preserving device longevity, power, and thermals.
• Some view local/self-hosted AI as philosophically similar to open source and as a check on concentrated corporate control.

WSL vs Native Linux on Surface

• Some argue the “trick” with Surface devices is to use WSL rather than native Linux, since hardware support (touch, power, odd peripherals) is incomplete.
• Others object that WSL lacks multi‑touch/gestures, has slow I/O / high CPU (vmmem), and still forces use of Windows’ disliked UX.
• There’s skepticism about running Linux on Surface at all, but also pushback: the linux‑surface project is described as close to upstream with a modest patch set and active reverse‑engineering work.

Surface Hardware: Performance, Reliability, Value

• The m3‑6Y30 SP4 in the review is widely seen as underpowered (very low TDP, tiny GPU), and 4 GB RAM is described as marginal in 2024.
• Reports of SP4/7 instability: ghost/failed touchscreens, flaky Wi‑Fi/Bluetooth, short battery life, poor thermals.
• Newer SP8/9 and ARM-based models get more positive experiences (usable coding, pen, travel, good battery on ARM), but are considered overpriced for their specs.
• Some like the form factor (3:2, pen, detachable keyboard) and use Surfaces as primary machines; others find them slow, fragile, or “weird tablets pretending to be laptops.”

Linux on Laptops/Tablets: Touch, Power, and Distro Choices

• Touch on Linux is described as “mostly works but janky”: Firefox touch issues, occasional breakage, poor tablet‑only install experience, reliance on USB keyboards.
• Several report good results on specific hardware: SP5 with NixOS, SP7 with Fedora, SP2 with Fedora, Surface Laptop 4 with EndeavourOS, Dell XPS 13 “Developer Edition” with Ubuntu, Framework 13 with Fedora.
• Others describe the classic Linux desktop pattern: initial excitement → driver/power issues (sleep, Wi‑Fi, Bluetooth) → abandonment back to Windows/macOS.
• Arch/Arch‑based distros (EndeavourOS, Manjaro) are praised as lighter and smoother than Fedora/Ubuntu on low‑RAM devices, largely due to different defaults (zswap vs zram, less background services).

Swap, zram, and Memory Management Debate

• The article’s disabling of swap/zram and subsequent OOM sparks a long debate.
• One camp: swap/zram is essential on 4 GB systems and generally beneficial on workstations; modern zram + MGLRU can keep systems responsive and extend disk cache.
• Opposing views: swapping (even to zram) often leads to unresponsive systems; better to disable swap and rely on OOM killers, especially for large numeric/scientific workloads where swaps indicate a bug or mis‑sizing.
• Multiple tuning recipes are shared (zram + LZ4, aggressive swappiness, sysctl tweaks), but commenters agree guidance online is often outdated and inconsistent.

Battery Life, Sleep, and “Modern Standby”

• Many complain that only Apple reliably delivers excellent battery life and sleep/wake behavior; x86 laptops on both Windows and Linux often have poor standby and wake‑in‑bag issues.
• Windows “Modern Standby” is heavily criticized: systems may run substantial background work while “asleep,” heating bags and draining batteries; S3 “traditional sleep” support is being removed.
• Some report Mac‑like reliability with carefully chosen Linux hardware (e.g., XPS Developer Edition, certain ThinkPads/Ideapads with tuned distros); others still see flaky sleep, Wi‑Fi crashes after suspend, or need to rely on hibernate.

Alternatives, Expectations, and Form Factor Trade‑offs

• Several emphasize that “Linux support” is primarily a hardware vendor choice: Linux‑first laptops (Framework, System76, Starlabs, Dell Dev Editions) integrate far better than random Windows machines.
• 2‑in‑1s and tablets: mixed views. Some love Surfaces, Galaxy Books, and Yogas for pen input, note‑taking and travel; others find them compromised as both laptops (ergonomics, keyboards) and tablets (weight, battery).
• Alternatives mentioned for Linux‑friendly tablet‑like use include Steam Deck, Minisforum V3, and Starlabs’ Starlite.
• Broader theme: if you want a smooth Linux experience, buy hardware designed and tested for it; expecting flawless Linux on proprietary tablets like Surface remains risky.