Hacker News, Distilled

Headline Number and 7% Assumption

• Many commenters find the "$400k household income" claim implausible on its face, since most families use childcare without earning anything close to that.
• Others point out the article’s math: ~$28k/year for two kids and an HHS “affordable” benchmark of 7% of income → ~$400k needed to keep costs under that threshold.
• Several note that the shock value comes almost entirely from the 7% rule; most families simply spend more than that share of income on childcare.

Actual Childcare Costs & Economics

• Numbers around $1.5–2k/month per kid in many US locales are reported, with wide regional variation and higher costs for infants due to stricter ratios.
• Some argue $28k/year for infant + 4‑year‑old is a realistic national average; others say even that feels absurd relative to many workers’ wages.
• Explanations include labor intensity, legally mandated child-to-caregiver ratios, high overhead, and Baumol’s cost disease (low-scope for automation).

Role of Regulation and Informal Care

• Several recall an older norm of neighborhood moms informally watching multiple kids cheaply.
• Others say modern licensing, inspections, and ratio rules make small-scale, informal paid care effectively illegal or uneconomical, pushing families toward expensive centers.
• Some defend regulation as necessary to prevent neglectful or abusive “cheap care” and to avoid long-run social costs.

Capitalism, Taxes, and Subsidies

• Thread branches into debates about capitalism vs “state capitalism,” tax levels on the rich, and whether taxing wealth would meaningfully change childcare costs.
• Some argue childcare is exactly the kind of sector that should be heavily subsidized as social infrastructure; others say taxes and redistribution can’t solve structural cost drivers.

International Comparisons

• Multiple commenters contrast the US with Europe/Scandinavia/Canada: subsidized or income-based fees, low out-of-pocket costs, long protected parental leave, and a political willingness to fund this via taxation.
• Counterpoints note that even in Europe, high-end or private care can still reach US-like prices and capacity remains a problem in some places.

Family Structure and Gender Roles

• There is substantial discussion of stay-at-home parenting vs dual-income households, with some praising “traditional” single-earner families as both cheaper and higher-quality care.
• Others highlight economic reality (mortgages needing two incomes), single-parent households, and equity concerns: long career breaks disproportionately harm women.
• Several comment on the emotional side: many mothers reportedly want to spend more time at home; some fathers feel less innate interest in babies, prompting nature vs nurture arguments.

Broader Social Tradeoffs

• A recurring theme is that lack of leave, universal healthcare, and subsidized childcare are themselves costly—Americans “pay” via stress, debt, and foregone family time.
• Some personal stories show families choosing one parent at home and retiring earlier; others spend six figures on childcare so both parents can work, then question if the tradeoff was worth it.

Code vs Software: Cheap Typing, Expensive Thinking

• Many argue that “writing code” has been cheap for years; what’s costly is designing good software, understanding domains, and making correct tradeoffs.
• LLMs mainly remove the time‑consuming “typing” step, not the hard parts: requirements, architecture, edge cases, performance, security.
• Several compare this to outsourcing or McDonalds: it’s always been easy to get lots of mediocre output cheaply; high‑quality work still costs.

Maintenance, Liability, and Technical Debt

• Strong consensus that code is a liability: more LOC means more to understand, test, and maintain, regardless of who wrote it.
• AI accelerates creation of “vibeslop” and prototypes that “sort of work” but are hard to change; reviewers and maintainers bear a growing burden.
• Some foresee more outages and new kinds of risk as agents rewrite critical systems; others think AI will eventually help with maintenance too.
• Reading/understanding code remains as expensive as ever, maybe more so when AI produces verbose, over‑abstracted structures.

AI as Autopilot: Where Humans Still Matter

• The pilot/autopilot analogy recurs: AI can fly the happy path (CRUD apps, boilerplate, simple glue code), but humans are needed for messy realities and emergencies.
• Good outcomes require tight human steering: clear specs, tests (often TDD), adversarial review, and careful use of agents rather than blind trust.
• A key new skill is “directing cheap inputs”: using agents to rapidly try approaches, then judging which won’t explode later.

Prototypes, Throwaway Code, and System Design

• Cheap code makes multi‑prototype exploration viable: build three versions in a day and pick one.
• Others warn that if you never personally wrestle with the problem, you don’t develop taste or understanding; the process, not the artifacts, teaches you.
• Some advocate “tracer bullets” and eval‑driven development: quick, end‑to‑end slices that are intended to be kept, not disposable demos.

Jobs, Skills, and Market Dynamics

• Commenters expect fewer roles for “ditch digger” programmers who just translate specs to code; top‑tier designers/architects remain in demand.
• Bootcamp‑style “I can type the code” skills are seen as commoditized; domain understanding, system thinking, and reliability judgment are the real moats.
• Debate continues on whether productivity gains are visible yet and how much AI will compress demand for mid‑level developers.

Transistor scaling and node naming

• Commenters clarify that the news is about throughput per EUV tool, not shrinking transistors.
• Modern “3nm” processes still have gate pitches and widths in the 30–50 nm range; some features go down to ~10–14 nm, showing how marketing diverges from physics.
• There’s discussion of moving from “nm” to angstrom labels (e.g., 18Å) and jokes about “0 nm” / “-1 nm” being pure marketing.
• Several people argue a better metric would be average gates per mm² rather than node labels, likening current terminology to loose “free range” food labels.

EUV light source mechanism and upgrades

• The core method: spray microscopic tin droplets in vacuum and hit each with precisely timed laser pulses to create a tiny EUV-emitting plasma.
• The disclosed advance: doubling droplet rate to ~100,000/s and moving from one to two shaping pulses (plus the main pulse), pushing source power from ~600 W to 1,000 W, with a roadmap to 1,500–2,000 W.
• Commenters stress how extreme this is for a vacuum system highly sensitive to heat and contamination.

Why EUV is uniquely hard

• Multiple explanations contrast EUV with X-rays and visible light:
  • Many materials are opaque at these wavelengths, forcing all-mirror optics with tight reflectivity and absorption constraints.
  • X-ray tubes are efficient only at much higher voltages (shorter wavelengths); at EUV energies they’d be absurdly inefficient and thermally impossible.
  • Focusing and photoresist behavior get much harder at higher photon energies; X-ray lithography exists but is considered even less practical and more stochastic than EUV.
• Overall consensus: EUV “barely works” and required a moonshot-level effort.

Engineering complexity and cleanliness

• Interior conditions must be far cleaner than any human cleanroom while continuously “exploding” tin; this drives extreme purity, maintenance, and multi-hundred-million-euro tool costs.

Corporate, geopolitical, and competitive context

• Thread pushes back on framing this as ASML vs “U.S. rivals”: the EUV source is developed by a U.S. subsidiary, within a broader multinational supply chain (optics, mechanics, etc.).
• Some note U.S. funding and deliberate tech transfer; others emphasize ASML as a genuine European-led integrator of global technologies.
• Potential competitors mentioned include a U.S. light-source startup backed by CHIPS Act funding and some Japanese research efforts, but timelines are seen as long and uncertain.

Economic impact and AI demand

• Several commenters predict that increased throughput will mainly feed AI accelerators.
• Frustration that consumer CPUs/GPUs are delayed or capacity-constrained due to AI demand, with one call for heavy-handed regulation to prevent AI firms from monopolizing advanced manufacturing.
• Others skeptically note that even with more logic chips, memory and storage capacity may remain bottlenecks.

Rust rewrites vs. decades of bug fixes

• Several commenters like Rust-based tools (e.g., ripgrep, fzf) but worry that 20–40 years of hard‑won bug fixes and quirky behavior in C utilities can’t be replicated quickly.
• GNU coreutils is seen as extremely stable; rewrites must be bug‑compatible, not just standards‑compatible, or they will break real‑world scripts.
• Example: reports that rust-coreutils dd breaks Makeself/CUDA installers highlight how subtle behavior differences can matter at distro scale.

Rust maturity, safety, and “unsafe”

• Some argue Rust is no longer “new hotness” — it’s over a decade old and widely deployed (Linux drivers, automotive, etc.).
• Others point out that Rust’s safety gains rely on avoiding or carefully fencing unsafe; system-level work often needs unsafe, weakening the “rewrite = automatically safer” narrative.
• There is also concern about projects rewriting well‑working tools “for virtue signaling” rather than clear technical need.

Dynamic linking, ABI, and large systems

• A long subthread debates Rust’s lack of a safe, stable native ABI.
• Current Rust interop uses the C ABI, which is as unsafe as C itself and doesn’t expose Rust’s richer type system at dynamic boundaries.
• Critics say this limits Rust’s usefulness for very large, dynamically linked systems and leads to code and stdlib duplication when many Rust .sos are used.
• Others counter that C‑ABI + Rust internals is still a major improvement and that fully stable ABIs historically freeze languages (C++ STL).

Licensing, GPL vs MIT, and TiVoization fears

• Multiple comments object less to Rust and more to Ubuntu adopting MIT‑licensed core utilities instead of GPL ones.
• Fear: a permissive userland enables locked‑down, non‑modifiable Linux systems (TiVoization-style), especially when combined with secure boot, attestation, and systemd‑centric tooling.
• Some see this as replacing “pro‑user” GPL software with “pro‑business” equivalents; others argue MIT still keeps original code open and avoids GPL adoption barriers.

Ubuntu/Canonical trust and distro politics

• Canonical is criticized for a history of pushing immature tech (PulseAudio early, snaps, Mir, sudo‑rs/rust‑coreutils) into users’ default path.
• Concern that Ubuntu may create a semi‑incompatible Rust‑based userland that fragments the Linux ecosystem.
• Several commenters say they’ve already switched to Debian, Mint, Fedora, etc., and urge Ubuntu‑derived distros to reconsider their base.

Rust ecosystem, stdlib size, and AI

• Some feel Rust’s ecosystem is still immature for less common domains, with many pre‑1.0 crates and API churn risks.
• There's debate over Rust’s intentionally small standard library: some want a .NET‑style rich stdlib; others prefer a “blessed crates” layer instead of freezing too much in std.
• AI tooling is reported to make Rust more approachable: strict types, good error messages, and compile‑time checking allegedly help agents iterate Rust code to correctness more reliably than dynamic languages.

Study design and causation vs. correlation

• Many commenters think the NPR framing overreaches: the underlying paper is about correlation, not proven causation.
• Major concern: excluding teens with diagnosed mental illness doesn’t exclude those already symptomatic but undiagnosed.
• Multiple people argue that only a randomized, blinded trial (cannabis vs. placebo) could really establish causality, and that such a trial is effectively impossible/unethical in teens.
• Some note that cannabis use in teens is still a marker of social “deviance,” making it hard to untangle drug effects from background risk factors.

Predisposition, self‑medication, and confounders

• Common alternative explanation: teens predisposed to mental illness may be more likely to use cannabis, nicotine, other drugs, or engage in other risky behaviors as a form of self‑medication.
• Cigarette smoking’s strong correlation with schizophrenia is cited as a non‑causal analogy.
• Others stress that mental illness, family stress, poverty, trauma, and broader behavioral patterns are all potential confounders that aren’t fully controlled.

Risk size, absolute vs. relative

• Some highlight the reported “2x risk” as large and concerning.
• Others emphasize base rates: ~0.8–1% of the sample developed serious disorders, so the absolute increase in risk may be around 1 percentage point, which could still be compatible with self‑selection.
• A long comment criticizes media and researchers for blurring relative vs. absolute risk and treating triggers in susceptible people as if they create illness in everyone.

Potency, dosage, and modern cannabis

• Several note that today’s cannabis (concentrates, high‑THC strains, edibles) is far more potent, likened to selling only very strong liquor.
• Some describe “micro‑dose” style use (low‑mg edibles, low‑THC vape pens) as a safer, underpromoted pattern vs. heavy daily use.
• Others argue that stronger products don’t automatically mean higher harm if people actually use less—countered by claims that many don’t.

Anecdotes and lived experience

• Multiple anecdotes describe cannabis‑associated psychosis or long‑term cognitive dulling, especially with heavy or early use, convincing some never to touch it.
• Others point out strong survivor bias and self‑selection in such stories and insist that millions use cannabis without severe issues.
• There is broad agreement that cannabis can cause paranoia and acute mental deterioration in some, especially vulnerable individuals.

Legalization, age limits, and politics

• Several argue that even if teen use is risky, that mainly supports 18/21+ age limits, not prohibition.
• Others are firmly against legalization, citing observed harms and what they see as denial and “motivated reasoning” among some pro‑cannabis advocates.
• Opponents are challenged on consistency (alcohol and tobacco policy) and on relying too heavily on anecdotes.
• Some commenters note that both pro‑ and anti‑legalization camps have used exaggerated or misleading claims, which erodes trust and polarizes discussion.

Open questions and suggested research

• Commenters ask about plausible biological mechanisms but note these remain unclear in the thread.
• Suggestions include using legalization as a “natural experiment” (e.g., comparing mental health trends across states over time).
• Overall, many want better‑designed longitudinal studies that more carefully handle predispositions, self‑medication, and social context before making strong causal claims.

Talent Shortage vs Cheap Labor

• Strong disagreement over whether offshoring and H1B use are about “lack of US talent” or simply “labor at salaries companies want to pay.”
• Some argue many roles (systems, OS, eBPF, security) genuinely require deep CS fundamentals, making it uneconomical to pay $150k+ to train juniors from scratch.
• Others counter there is plenty of qualified domestic talent; companies just prefer cheaper offshore or visa-dependent workers and then justify it with “pipeline” rhetoric.
• H1B is described by some as “modern indentured servitude” due to visa dependency and layoff risk; others who held visas say they felt grateful, not exploited.

Corporate Incentives and Shareholder Primacy

• Repeated theme: public companies are loyal to shareholders, not to US workers or national interests.
• Critics trace much of this to legal and cultural “shareholder value” doctrine (e.g., Dodge v. Ford, buybacks), which strongly pushes wage minimization and offshoring.
• Some argue this is exactly how free markets work—capital moves to cheaper labor unless regulation forbids it. Others say that’s a design flaw, not a neutral fact.

H1B Restrictions and Offshoring

• Many note that layoffs and offshoring predate recent H1B “scrutiny,” but agree that tighter visas accelerate opening offices in India/Eastern Europe/elsewhere.
• Viewpoints split:
  • One side laments the US “throwing away” its advantage as world talent relocates or is retained abroad.
  • Another welcomes Silicon Valley’s relative decline and questions why US workers should be protected over equally capable foreign workers.

Impact on India and Other Hubs

• General consensus that India benefits: higher demand and pay for engineers, stronger ecosystem, and potential for more entrepreneurship as returnees bring organizational experience.
• Some note downside for Indian startups facing rising salary competition from multinationals.
• Israel is cited as an example where salaries are not dramatically lower than US secondary cities, yet companies feel they get stronger fundamentals (OS, algorithms).

Cycles, Local Economies, and Policy

• Comparisons to early-2000s offshoring: initial savings, quality problems, hollowed domestic pipeline, then partial reversal.
• Several comments describe a broader cycle: offshoring depresses local wages and demand, cities decline, then “onshoring” is rediscovered as visionary.
• Proposed responses range from strict protectionism to withdrawing tax breaks and contracts from heavy offshorers, to comprehensive reforms of labor, corporate governance, and social safety nets.

Version Control & Git Education

• Many welcome the strong version-control chapter, arguing good commit history and use of tools like bisect/blame/rebase dramatically improve debugging and collaboration.
• Others note most developers learn only minimal git workflow at work or “by necessity”, leading to poor histories and cargo-cult usage (delete/re-clone when confused).
• Disagreement over responsibility: some say widespread misuse means git is badly designed (unintuitive, jargon-heavy, poor “undo”); others see it as a powerful tool that simply requires training, like a bandsaw or a Bloomberg terminal.
• Alternatives (Mercurial, jj, GUI frontends, git aliases/scripts) are praised as higher-level or friendlier interfaces over git’s “assembly-like” CLI.
• There’s debate over whether clean commits should matter in corporate settings where the PR (often squash-merged) is the real unit of work.

Scope and Value of the “Missing Semester”

• Many consider this kind of course one of the most useful in their education, solving day-to-day blockers (shell, git, tooling) that fundamentals don’t address.
• Some report similar 1-credit or UNIX-tools courses at their universities and say they still rely on those notes. Others note departments often wanted but weren’t allowed to teach such “non-academic” skills.
• Suggested additions:
  • Practical IT skills: information management, backups, self-hosting basic services, basic troubleshooting.
  • Tools: sed/awk, shell mastery, scripting, statistics/data tools (Polars/Plotly), LaTeX/org/R/Quarto, touch typing.
  • Software topics: testing/QA (possibly its own course), deeper software quality (complexity, maintainability, modularity).
  • “Beyond code”: interaction with OSS communities, interviewing, salary negotiation, team leadership, communication with management, career progression, and personal hygiene.

AI, Agentic Coding & CS Degrees

• Course authors explicitly ask about including AI topics. Some commenters support this and want more, e.g., a “build your own agent” lecture as high-leverage practice.
• Others see “agentic coding” as hype that doesn’t deserve space, suggesting the course should focus on understanding systems, not operating AI tools.
• Broader debate: is CS education still worthwhile when AI can write code?
  • One side: for people viewing CS purely as vocational “coding”, maybe less so.
  • Counterarguments: CS is distinct from coding; LLMs excel at recombining known patterns but struggle with novel formulations and poorly documented domains; human insight remains essential.
  • Concerns that many will become low-paid “LLM operators” vs higher-value software engineers.

Comments, Ethics & Soft Skills

• The “Beyond the Code” section on comments is appreciated; guidance that comments explain “why” rather than restating “what” resonates.
• Discussion around TODOs: they often rot unless tied to tickets or tagged with initials to preserve intent.
• Some are surprised ethics isn’t more central as a “missing semester” topic, though others question how much a short course can shift moral compasses.

Identity, Moderation, and AI

• Some argue federated systems are bottlenecked by identity and moderation; without strong identity, spam and abuse are hard to manage.
• Others counter that AI agents are eroding identity as a reliable signal anyway; systems may have to judge content/actions instead of who posted.
• One concern: AI “launders” human ideas and hides the operator’s intent, making some users prefer original human expression over agent output.
• A radical view suggests accepting Sybil/bot swarms and designing systems where uniqueness has no value and only originality matters.

Hosting, Ownership, and Infrastructure

• Strong irony is noted: arguing for independence while hosting on a big corporate static hosting platform.
• Defenders say the key is owning content and domains; using corporate hosting as a replaceable convenience is acceptable if it’s portable.
• Critics reply that using a corporate subdomain weakens that claim, and many non‑giant alternatives exist.
• Home hosting raises issues: IP exposure, dynamic addresses, NAT, ISP hostility to servers, security of always-on devices, and the burden of fighting bots.
• Some propose tunnels/relays (overlay networks, reverse proxies) as a compromise, though others note that if you already pay for a VPS, simple hosting there may be easier.

Usability and Tooling

• Broad agreement that “simple to use” is the missing piece. Pi OS, Docker, and current homelab setups are seen as far beyond most people.
• Markdown itself is argued to be too technical; WYSIWYG editors, Word-style tooling, or folder-based CMSs are preferred.
• Several projects aim to simplify self-hosting (personal app platforms, git-backed CMSs, mobile apps that publish to static hosts), but all still have rough onboarding.
• Some see AI as a potential UX layer: users describe what they want and get custom apps/sites, though others doubt most people can even articulate requirements.

Discovery, Attention, and Incentives

• Many say publishing isn’t the real bottleneck; discovery, attention, and monetization are.
• People post to big platforms because that’s where friends and audiences already are, and because they get instant feedback (likes, comments).
• Personal sites often feel like “throwing work in the trash” without a discovery and interaction layer.
• Suggestions include reviving RSS with better directories, or building decentralized search, but scalability and spam/SEO-like gaming are concerns.

Structural Limits and Alternative Visions

• Several comments stress that ownership hits a wall at ISPs and backbone cables; control of physical infrastructure and state censorship remain fundamental constraints.
• Co-ops and new protocols (P2P networks, alternative naming/identity systems, mesh networks, overlay “new internets”) are seen as promising but politically and economically hard to scale.
• Some see the article as inspiring but technically and socially naive: running small static pages is easy; replacing today’s app- and video-centric, attention-driven web is the hard unsolved part.

Technical capability & memorization

• Some argue it’s unsurprising: if models are next-token predictors, any novel in the training set is just one valid token sequence, so there must exist prompts that elicit it.
• Others counter that predicting an unseen novel verbatim is astronomically unlikely; being able to do so strongly implies the text was in training data.
• Several commenters emphasize LLMs as lossy compressors (pigeonhole principle), not perfect archives; verbatim output probability depends on how often and how redundantly a string appeared during training.
• Reported results include long “near-verbatim blocks” of thousands of tokens from famous novels, not just single-sentence completions.

Significance of the results

• One camp calls this a “nothingburger”: 70% sentence-level match and imperfect runs mean you still need the original to reconstruct a clean book.
• Others think it’s significant legally and evidentially: being able to extract 70%+ or multi‑thousand‑token continuous chunks is likely enough to prove inclusion in training data and to interest litigators.
• There’s interest in whether less-popular, sparsely quoted books can be similarly reproduced; that would be more worrying.

Copyright: is training a “copy”?

• Strong view: the violation occurs at copying into the training set, regardless of later output or transformation. If the model weights encode works that can be reproduced, they contain copies.
• Counter-view: models are more akin to humans who read and “learn” from books; the key legal issue should be downstream distribution, not mere internalization.
• Disagreement over fair use: some think training will ultimately be justified as transformative; others think copyright law’s text (e.g., US “copy” definition) clearly covers model weights.

Guardrails, jailbreaks, and liability

• Paper notes some models require jailbreaking to extract text; others comply with simple continuation prompts.
• Debate over whether needing jailbreaks counts as “circumventing a protection system” or just abusing a weak safety layer.
• Some argue liability should fall on the user who coerces the model into infringement; others say providers are responsible if their product readily enables mass reproduction.

Human vs machine analogies

• Frequent comparisons to humans memorizing books, singing songs, or writing fanfic; critics respond that:
  • Computers are explicitly covered as “machines/devices” in copyright law, unlike human memory.
  • Human-scale memorization/distribution is rare and low‑impact; LLMs scale to millions of perfect or near‑perfect copies.
• General consensus: “humans also do this” is rhetorically appealing but legally weak.

Broader framings

• Some frame LLMs as super‑compressed libraries or search/index systems over the internet, now leaking underlying works.
• Others see this simply as large‑scale, automated plagiarism, enabled by messy, heavily lobbied copyright law that has repeatedly lagged technological shifts.

Hosting and Context

• Noted that the PDF is mirrored on an archiving-focused personal site that hosts many documents; some discuss that site owner’s broader work on web archiving.
• One commenter pasted a biographical summary (career, Fields Medal, early SAT score) to “save a click”.

Emotional Reactions & Literary Analogies

• Several readers misread the title as an obituary and briefly panicked.
• Many found the report deeply moving or humbling, comparing it to Flowers for Algernon in how it tracks development through written work.
• Others highlight the sweetness of details showing him as a “happy little boy” who plays hide-and-seek and is accepted by older classmates.

Comparisons to Other Prodigies & Language Learning

• The childhood of a 19th‑century philosopher (Greek/Latin/Plato very young) is raised for comparison.
• Long subthread debates whether early multi‑language ability is actually remarkable, with many arguing that multilingual toddlers are common globally, but classical languages learned from books/tutors are a different kind of feat.
• Some are skeptical of hagiographic “speaks 7 languages” stories and “great man” mythology.

Nature, Nurture, and the Limits of Ability

• Big split: some argue that enough motivated practice (3–4 hours/day for years) could bring many children close to this level; others insist innate ability dominates at the extreme tail.
• Analogies to elite athletes and musicians are used on both sides: is “talent” just trained skill, or a real ceiling?
• Several mention observed limits: adults who fail basic calculus despite immense effort; others insist they’ve seen “talent” in memory, rhythm, etc. clearly outstrip effort.
• Intrinsic motivation, emotional stability, and supportive family environment are repeatedly cited as crucial.

Societal Value & Celebration of Gifts

• One thread asks why intellectual prodigies are morally “celebrated” while looks-based “natural” gifts (e.g., supermodels) are seen differently.
• Responses: we tend to value contributions with broader human impact; yet, in practice, beauty often gets more money and attention than math.
• Some argue everything—grit included—is “wired,” so all accomplishment is partly luck; others justify praise as a cultural choice that incentivizes socially valuable work.

Parenting, Schooling, and Support

• Multiple commenters credit the parents for making advanced materials and unusual schooling arrangements available, and not forcing drill.
• Others note even that “light” enabling work (materials, advocacy with schools, facilitating contact with mathematicians) is significant.
• Stories from readers: being bored and punished in school for being ahead; lack of enrichment for bright kids; idea of being or not being “school-shaped.”
• Discussion of accelerating him across grades in Australia; some note such cross‑year placement is rare or impossible in their countries.

AI, Future Intelligence & Benchmarks

• Some muse that extreme human prodigies show biological intelligence is nowhere near a genetic “max”; breeding or selecting for intelligence is discussed but questioned ethically and practically.
• Others suggest AI may soon outnumber or outperform top human mathematicians (“5 million Tao‑level agents”), raising comparisons to chess/Go.
• Skepticism about nurturing gifted kids with AI “companions”; human emotional connection is seen as more important.

Reflections on the Case Study Itself

• Readers love the BASIC program snippet with the whimsical “bye mr. fibonacci!” line; it evokes nostalgic memories of early programming and shows childlike humor alongside advanced math.
• Several appreciate test items with intentionally wrong/underspecified questions as a way to assess confidence and critical thinking; someone wonders if such adversarial questions are used in LLM benchmarks.
• Noted wording shift: the report’s phrase “meeting [his] special needs” now commonly implies disability, illustrating how educational language has changed.
• A few express discomfort and curiosity about whether the now-adult mathematician is okay with such an intimate childhood profile being public.

Information‑dense aesthetics

• Many commenters see Japanese web design as a continuation of pre-2010 “portal” style: dense, text-heavy, lots of options visible at once.
• Some praise it as clean but information-rich, reminiscent of paper catalogs, magazines, and older Western portals (Yahoo, Netscape, etc.).
• Others note similarity to physical Japan: drugstores, signage, pachinko parlors, and magazines that are visually busy and packed with text.
• Contrast is drawn with Chinese sites: also maximalist, but often more animated and gamified (popups, confetti), whereas Japanese pages are described as more static and utilitarian.

Usability: efficient vs overwhelming

• Fans claim Japanese pages respect users’ intelligence, prioritize function, and reduce scrolling; everything is visible instead of hidden in hamburgers/three-dot menus.
• Suggested alternatives to hidden menus: bottom tab bars, contextual toolbars, right-click menus, classic desktop menu bars, and higher information density on desktop.
• Critics find many Japanese sites (especially e-commerce, government, and travel) convoluted, confusing, and inconsistent (clickable vs non-clickable images, weak hierarchy).
• Several people living in Japan say they avoid local shopping sites where possible, calling flows “hostile” despite the information density.

Technical and historical factors

• Legacy constraints are cited as major drivers: early CJK encoding issues, limited fonts, difficulty expressing typographic hierarchy, and reliance on images for text.
• Examples of dated internal tools (framesets, IE-era design) are framed as continuity rather than ignorance: systems work, users are trained, so there’s little incentive to redesign.
• Persistent practice of scheduled nightly downtime and batch processing (e.g., rail passes, transit cards, games) surprises users used to 24/7 availability.

Cultural context and stereotypes

• Commenters link layouts to Japanese print traditions (multi-directional text, newspaper-like columns) and high literacy, enabling more compressed information.
• Others highlight Japan’s partial cultural and linguistic separation: many users mainly consume domestic content, so Western design trends diffuse slowly.
• There’s pushback on the stereotype of “Japanese minimalism”: people describe a spectrum ranging from extremely minimalist (e.g., certain brands) to extremely maximalist.

Minimalism fatigue and global trends

• Several participants express fatigue with Western “corporate minimalism”: giant hero images, low information density, excessive whitespace, and endless scrolling.
• Minimalist design is seen as signaling “luxury” and high-end positioning, while dense designs communicate bargains or straightforward utility.
• Some note that Japanese web design is evolving and that the article’s broad cultural conclusions may already be less accurate.

Responsibility for children’s access

• Strong split between “parents should manage kids’ devices and behavior” vs “platforms and states must gate access.”
• Many argue the internet should be treated like alcohol or cigarettes: adults can buy, but supplying minors is regulated and punishable.
• Others counter that most parents are overwhelmed, under‑informed about controls, or themselves digitally addicted; relying on parenting alone is unrealistic.

Age checks, ID, and de‑anonymization

• Repeated concern that “age verification” is actually an identity system: once you prove age with government ID, platforms, governments, and data brokers can eventually link accounts to real people.
• Critics say that if child safety were the true goal, laws would target addictive design (infinite scroll, recommendation algorithms), not identity collection.
• Device‑ or browser‑level “I am a child” flags and site self‑rating are proposed as alternatives that don’t require IDs, but doubters say bad actors simply won’t flag themselves.

Technical proposals and limits

• Suggested architectures:
  • Device‑side age flags passed in HTTP headers or via OS APIs.
  • Government- or bank‑issued digital credentials with zero‑knowledge proofs (prove “over 18” without revealing identity).
  • Token systems bought in person after ID check, then used anonymously online.
• Pushback: any usable system must prevent large‑scale sharing and re‑use of tokens or credentials, which tends to reintroduce tracking, rate‑limits, revocation lists, or hardware attestation.
• Several note that “perfect” cryptographic systems are complex, hard to deploy, and will be bypassed in favor of simpler, more invasive vendors.

Effectiveness and workarounds

• Many argue age‑gating will stop honest users but not determined kids:
  • Borrowing parents’ or older siblings’ devices/IDs.
  • Using school devices, public Wi‑Fi, VPNs, Tor, foreign sites.
• Analogy: like underage drinking—laws reduce use, don’t eliminate it. Some say “imperfect but better than nothing”; others call it mere security theater plus privacy loss.

Government power and surveillance concerns

• Strong undercurrent that this is part of a broader push to de‑anonymize and control online speech, using “protect the children” as pretext.
• Fears include:
  • Linking real‑ID to all social media for political repression and chilling dissent.
  • Expanding device attestation, banning rooted/jailbroken systems, and effectively killing general‑purpose computing and anonymous browsing.
• Some see coordinated lobbying by age‑verification vendors and large platforms who benefit from verified, targetable users.

Alternatives and tradeoffs

• Proposals emphasize:
  • Strengthening and simplifying parental controls at OS/router level.
  • Regulating social media design (addiction mechanics, targeting children) and corporate incentives, rather than identity.
  • Accepting an imperfect, more anonymous internet vs a “safer” but tracked and permissioned one.

Verified Test Results (What VTT Actually Confirmed)

• Cell under test: 26 Ah, ~94 Wh, 3.6 V nominal, 2.7–4.15 V recommended range (4.3 V max).
• Fast charging performance:
  • 5C (130 A): 0–80% in ~9.5 min, 0–100% in ~13.5 min.
  • 11C (286 A): 0–80% in ~4.9 min, 0–100% in ~7.3 min.
• Charge/discharge: delivered 98.4–99.6% of the charged capacity by Ah even after 11C, but only ~90% round‑trip energy efficiency by Wh.
• Thermal behavior with heatsinks:
  • Two-sided cooling: ~47°C at 5C, ~63°C at 11C.
  • One-sided: up to ~61.5°C at 5C and ~89°C at 11C (still functional but near limits).

What’s Missing / Open Questions

• No weight or volume → no independent energy density figure.
• Only 7 cycles run → no evidence on cycle life or 100k‑cycle claims.
• No data on cost, materials, “non-hazardous” or “non-geopolitical” claims.
• No abuse tests (nail, crush, short, overcharge) or extreme temperature performance.
• No data on self‑discharge or long-term charged storage.

Status and Role of VTT

• PDF is digitally signed; multiple commenters trace it to an official VTT contact and a VTT press release, and assume the report itself is genuine.
• VTT is described as a government-owned, non-profit test lab that runs whatever tests the client specifies; it is not an independent auditor of all product claims.
• Critical nuance: VTT states it tested “devices supplied by the customer, which the customer identified as solid-state battery cells” – it did not verify chemistry or composition.
• Commenters note VTT did not record cell weight/dimensions and cannot guarantee that identical cells are used across different test campaigns.

Hype, Red Flags, and Scam Debate

• Many are excited that performance is at least comparable to good Li‑ion and that fast charging with relatively simple cooling is independently measured.
• Others argue these C‑rates are achievable with existing chemistries (e.g. FPV LiPos, BYD’s fast‑charging LFP), so the test doesn’t prove a breakthrough.
• Significant skepticism around the company’s broader behavior: drip‑marketing of partial results, founder’s previous “magic AI/ASI” and trading products, a related motorcycle company with troubling audit findings, and rumors of loosely regulated fundraising in Finland.
• Some call it an obvious scam; others push back, noting failed or hype-driven startups aren’t automatically fraudulent and demanding concrete evidence of harmed investors/customers.

Vehicles and Commercialization Timeline

• Existing Verge motorcycles reportedly use conventional ~20 kWh Li‑ion packs; no bikes with the claimed solid-state pack are in customers’ hands yet.
• Solid-state-equipped models are promised for 2026; early “available today” messaging contrasted with later statements about first deliveries months later, which commenters see as a key timeline to watch.

Broader Context and Outlook

• Commenters stress that many major players already have working solid-state prototypes; the real challenge is scalable, cheap production with good density, cycle life, and safety.
• Several plan to reserve judgment until independent groups, not arranged by the company, can acquire cells and run full characterization (energy density, cycle life, abuse testing) end-to-end.

Scope and Structure of Hetzner’s Price Increases

• Increases affect both cloud (VPS) and dedicated servers, for new and existing customers from 1 April 2026.
• Indicative ranges from the thread:
  • Cloud VMs: ~36–38% increase.
  • Bare metal: ~10–15% (many examples show +1–5 €/month).
  • Some very old/auction servers also rise slightly (cents to a few euros).
• A separate table shows massive RAM add‑on hikes (quoted as ~575% and “effective immediately”), but multiple commenters find inconsistencies and suspect errors in Hetzner’s published list.

Hardware Shortages and RAM Pricing

• Consensus that DRAM has gone up ~5x in ~6–12 months, with SSD/HDD prices also up and some parts “sold out” through the year.
• Several argue Hetzner is simply passing through sharply higher component costs; others note they had previously absorbed increases (energy, IPv4) but can’t anymore.
• Debate over RAM trajectory:
  • One side: prices have “stabilized at 5x” and will gradually fall as new capacity comes online.
  • Other side: manufacturers are not increasing non‑HBM capacity, so shortages and high prices may persist for years.

AI, Venture Capital, and Market Distortion

• Strong sentiment that AI hyperscalers are “vacuuming up” DRAM, SSDs, HDDs, and even wafers, using speculative VC money rather than sustainable profits.
• Some call this an “AI tax” on everyone else; proposals include special AI taxes or even rationing of components.
• Others counter that this is textbook demand shock: prices reflect genuine (if bubble‑driven) demand, not classic manipulation.
• Disagreement over whether this is a cyclical spike or a structural shift that permanently hands consumer/SMB hardware markets to Chinese manufacturers.

Comparisons and Alternatives

• Even after increases, many say Hetzner remains far cheaper than AWS/GCP/DO for equivalent specs; some note DigitalOcean is vastly more expensive at similar RAM/disk.
• OVH is also raising prices, in some cases more aggressively. Other EU options mentioned: Netcup, Scaleway, Contabo, Seeweb, Leaseweb; all expected to face similar pressure.
• Some users plan to:
  • Lock in/add extra dedicated servers now.
  • Move tiny workloads to home servers, old PCs, or Raspberry Pis (with cautions about power costs, ISP terms, and insurance).

EU vs US Clouds and Policy Backdrop

• Part of Hetzner’s growth is attributed to European customers wanting to avoid US‑based clouds (Cloud Act, perceived political instability, tariffs).
• Discussion that Europe is still dependent on non‑EU DRAM/CPU supply and lacks strong domestic memory fabs, limiting its ability to shield itself from AI‑driven shortages.

Impact on Developers and Software Practices

• Concern that higher entry‑level VPS prices hurt hobby projects, indie SaaS, and small startups built on sub‑10€/month boxes.
• Counter‑argument: if a few extra euros kill a startup, the business was too fragile; but several note side‑projects do die over exactly these recurring costs.
• Some see a “silver lining”: pressure to reduce bloat—less Electron, fewer oversized Kubernetes clusters, more efficient languages (Rust/Go), and a return to optimizing for limited RAM and storage.

Swift vs Rust vs C++ for Ladybird

• Ladybird previously tried Swift for memory safety but ran into real-world compiler/interop bugs (e.g. Swift failing to import some C++/Clang modules with no workaround) and weak non-Apple platform support.
• Rust–C++ interop is acknowledged to be basic (via C ABI, cxx.rs, Crubit, opaque pointers), but seen as reliable and battle‑tested (Firefox cited as example of mixed Rust/C++).
• Some are uneasy about language churn: Jakt → Swift → Rust feels volatile; others argue pre‑alpha is exactly when to pivot.
• Several comments note the choice of C++ in 2018 was understandable; Rust maturity and tooling are much stronger now.

Memory Safety and Language Debates

• Many argue browsers are the prime case for memory‑safe languages: they execute untrusted code and historically suffer from UAFs/overflows.
• Pro‑Rust side: “modern C++” plus guidelines still leaves many memory bugs; large C++ codebases (browsers, Android components) show measurable gains when pieces move to Rust.
• Skeptical side: modern C++ plus discipline can be safe; Rust is complex, slow to compile, and brings dependency bloat. Some advocate Go, Zig, D, or Ada/SPARK instead, but others point out Rust’s ecosystem, contributor pool, and explicit safety guarantees.
• Several note that partial Rust adoption (only some subsystems) doesn’t magically make the whole browser secure; the benefit depends on how much unsafe remains.

AI‑Assisted Port of LibJS

• LibJS lexer/parser/AST/bytecode generator (~25k LOC) were ported from C++ to Rust in ~2 weeks using Claude Code/Codex, guided by many small prompts.
• Hard requirement: byte‑for‑byte identical ASTs and generated bytecode between C++ and Rust, with zero regressions across ~65k tests (test262 etc).
• The Rust code intentionally mirrors C++ register allocation and structure, so it is correct but not idiomatic; cleanup is deferred until the C++ pipeline can be retired.
• Many see this “bug‑for‑bug compatible” rewrite plus strong tests as the right way to use LLMs; others worry about long‑term tech debt from non‑idiomatic AI‑translated Rust.

Rewrites, Focus, and Project Trajectory

• Classic “never rewrite” arguments are revisited: people note LLMs and strong test suites change the calculus, but warn that cleanup and future refactors can be as hard as a fresh rewrite.
• Some commenters think energy should go to making Ladybird a daily‑driver browser before large language migrations; others argue early adoption of a safer language is strategically smart and may attract more contributors.
• There’s broad agreement that AI here is being used in a disciplined, test‑driven, “assistant not replacement” role—but also concern about hype, “AI slop,” and the risk of endless yak‑shaving.

Site UX and Presentation

• Many found the site nearly unusable: scroll-jacking/autoscroll, snap-to-sections, and full-page clickable areas made it hard to read blurbs or open original HN links.
• Criticism focused on “hijacking scroll” being especially bad on mobile and even on desktop, with users unable to stop between sections.
• Some liked the mobile UX and aesthetics, but they were a minority voice.
• Several noted small quality issues (default favicon, awkward light/dark toggle) as reinforcing a “low-effort” or rushed feel.

Cherry-Picking, Nuance, and Survivorship Bias

• A major theme: the site is accused of cherry‑picking negative comments and presenting them as “HN didn’t love X,” while ignoring positive or nuanced replies in the same threads.
• Commenters stressed that any sufficiently large discussion will contain both praise and skepticism; you can build the opposite narrative just as easily.
• Multiple people called out survivorship bias: only successful outliers are shown, not the many similar ideas HN disliked that actually failed.
• Some asked for an “inverse list” of heavily praised HN darlings that went nowhere.

Definition of Success and “VC Lens”

• Many objected that the site equates success with valuation, acquisition, or funding, ignoring social costs or long‑term impact.
• Critiques of Uber, Airbnb, Bitcoin, LLMs, and React are seen as still valid even if those projects are now large or profitable.
• Some argue the page reads like a venture‑capital narrative: money made is treated as proof that early criticism was wrong. Others counter that markets can reward flawed or harmful products.

Specific Technologies and Products

• Tailwind and React: several say early HN criticisms remain accurate despite widespread adoption; popularity doesn’t prove technical or UX merit.
• DuckDuckGo: debate over whether the name hurt adoption; some think it’s silly and non‑verbable, others see it as no worse than “Google.”
• LLM tools and OpenClaw: many feel it’s too early to treat them as settled “wins”; early negative comments about quality and hype are described as still valid.

AI / Automation Concerns

• Several suspect the summaries/outcomes were generated by an LLM: repetitive style, oversimplified narratives, occasional factual overreach (e.g., Warp description later corrected).
• This contributes to a sense that the whole thing is a snarky, low‑nuance “HN was wrong” piece rather than a thoughtful retrospective.

Scope and Size of Price Increases

• Reported hikes of ~30–40% on many products, but some users see only ~3% on specific dedicated “server auction” machines.
• “Server auction” servers explicitly get a flat 3% rise.
• One example shared: €31.90 → €32.86 and €34.51 → €35.55 (both ~3%), suggesting the steepest increases hit other product lines.
• IPv4 addresses add an extra €0.60 on top of listed (IPv6-only) prices.

Official Justification and Communication

• Hetzner’s statement: large cost increases in infrastructure operations and hardware purchases; attempts to absorb costs have “reached the limit.”
• Some discussion around a clumsy translation (“IT branch” from German “IT‑Branche”), generally treated as a harmless language issue.
• A detailed price list exists on Hetzner Docs; staff confirm it covers both existing and new products starting 1 April 2026.
• Existing RAM add-ons may also be affected; separate emails reportedly sent.

What’s Driving Costs (According to Commenters)

• Hardware: RAM, storage, CPUs, and GPUs all cited as sharply more expensive, with shortages and long lead times.
• Energy: multiple people mention rising power and cooling costs, though one notes German electricity prices have recently fallen, so hardware is seen as the main driver.
• General view that “competing on price never lasts,” and Hetzner had been absorbing higher costs for years.

AI Boom, Bubbles, and Hardware Scarcity

• Strong thread linking price hikes to AI hardware demand: AI seen as absorbing massive RAM, GPU, and DC capacity.
• Split views:
  • One camp: AI is a bubble; unsustainable spend will collapse, and hardware prices will fall back.
  • Another camp: this is “the next industrial revolution”; AI makes senior engineers and creators vastly more productive, and prices for AI services will eventually rise, not fall.
• Debate over whether current AI usage is profitable or heavily subsidized, and whether growing cloud prices will curb demand.

Customer Impact and Reactions

• Some users say Hetzner remains vastly cheaper than AWS/Azure even after hikes.
• Others see this as a breach of trust, especially raising prices on existing infrastructure, and talk about moving storage or workloads to other low-cost providers (though competitors are also raising prices).
• A few are proactively buying second-hand servers and colo space to escape cloud pricing altogether, though others doubt 15–20 year hardware lifespans.

Use Case: Styrofoam vs. Cardboard

• Most commenters see this as a replacement for polystyrene/Styrofoam, not cardboard boxes.
• Mycelium packaging is framed as rigid cushioning inserts inside an outer cardboard box.
• Several note cardboard is already renewable and highly recyclable; plastics/foams are the main problem area.

Environmental Impact & Practicality

• Some doubt it’s “better than cardboard,” but agree it’s a strong alternative to plastic foam.
• Concerns:
  • Slow production (around a week to “grow” each piece).
  • Parts are relatively heavy and non-compressible, increasing storage and transport costs and emissions.
• Because of these constraints, people claim current adoption is mostly niche or marketing-driven for high-margin goods.
• Others remain optimistic but question whether it can ever be cheap enough to seriously displace plastic.

Competition, Claims & Geography

• Commenters list multiple European mycelium-packaging companies and question the “Europe’s first” claim.
• Debate over whether “Europe” vs “EU” vs “UK” is being used in a misleading way.
• Some note large brands have been using mycelium packaging for years via other suppliers.

Branding, Naming & PR

• The “Magical Mushroom” name is polarizing:
  • Some clicked specifically because it sounded like psychedelics.
  • Others think it hurts business credibility and corporate-sales “culture fit.”
• A few suspect coordinated PR/VC-driven promotion; others say it’s likely just organic virality.

Technical Properties & Inputs

• Product is explicitly positioned as polystyrene-replacement foam; performance claims match polystyrene’s.
• Weight and density vary widely based on recipe; higher density improves strength but increases weight.
• Fire safety is questioned; one link suggests a respectable fire rating, but details aren’t deeply discussed.
• “Agricultural byproducts” reportedly include fibrous hemp cores; users speculate about manure or woody waste.
• Questions about edibility receive answers that it’s compostable and biodegradable, not food.

Alternatives, Policy & Future Vision

• People compare mycelium to molded pulp, sugarcane, and corn-starch foams; unclear advantages besides novelty and possible premium feel.
• Some envision future “on-site grown” packaging at packing facilities, cutting shipping loops and enabling home composting.
• A few advocate regulation: phasing out plastics in favor of bioplastics and mycelium; others are skeptical of economic feasibility.

Mycology Tangent

• Thread diverges into hobby mushroom growing, substrates, contamination, and spore/health concerns, reflecting broader fascination with fungi and mycelium as a technology platform.

AI, “Frictionless Relationships,” and Artificial Intimacy

• Several comments reference Sherry Turkle’s ideas: AI creates “frictionless relationships” with “pretend empathy,” which people may prefer to real-world disagreement and negotiation.
• Some see this as addictive and worry chatbots are becoming substitutes for messy, real relationships.
• Others note this “more real than real” but low-friction world is a hallmark of postmodern life.

Confession, Power, and Data vs. AI Logs

• Parallel drawn between telling secrets to priests and to AI: both receive intimate disclosures, but one is bound by absolute secrecy, the other by ad-tech.
• Multiple commenters stress how seriously the Catholic seal of confession is taken (even to torture/death), contrasting that with Big Tech’s profit motive and data-mining.
• Some push back on framing confession primarily as “power”; they argue its point is liberation, while AI systems are explicitly about control and profit.

Homilies: Authenticity, Context, and Mediocrity

• Many note homilies are often recycled, phoned in, pulled from manuals, or overly political; AI would just be a new flavor of canned content.
• Others insist good preaching is community-specific: you can’t feed enough local, pastoral context into a model without breaching privacy or losing nuance.
• Some priests/seminarians in the thread report AI-generated homilies are “ok-but-not-great,” comparable to average sermons but lacking heart.
• There’s support for the Pope’s core point: the act of writing a homily is spiritually formative for the priest; outsourcing the thinking undermines that.

AI Dependence and Cognitive Atrophy

• Several commenters worry AI will atrophy critical thinking, making people dependent on subscription “thinking services,” likened to a new feudalism.
• Others note we’re already dependent on search engines and navigation; AI is just the next step.
• A recurring distinction: using AI for technical work (code, docs) is seen as acceptable; using it for spiritual guidance or intimate human communication feels wrong or disrespectful.

Religion, Science, and Institutional Smartness

• Long subthread debates the Church’s historical relationship with science (Galileo, crusades, scientific clergy) with substantial nuance and disagreement.
• Some are impressed by recent Vatican documents on AI, arguing the institution “gets it” intellectually, even if one rejects its theology.

Authenticity vs. AI-Mediated Speech

• Multiple commenters argue any important human-facing writing (sermons, emails, pastoral care) should be authentically human, even if clumsy.
• Using AI for these is described as “gross,” a betrayal of trust, and fundamentally missing the point of religious and interpersonal encounters.

Business model and monetization

• People are impressed by the tools but question how the company will make money, especially given VC backing and eventual pressure to “cash out.”
• Vite+ is described as the main monetization path:
  • Positioned as an all-in-one JS toolchain and monorepo solution targeting enterprises, competing with Nx/Moonrepo/Turborepo/Rush.
  • Value props: single configuration, shared AST/dependency graph info between tools, better caching and CI performance.
• Some see a tension: selling visual/enterprise tooling on top of a powerful open core might allow others to re‑implement similar experiences for free.
• Others argue enterprises prefer to pay money rather than spend developer time, so this is not a real paradox.

Rust vs JavaScript (and others) for tooling

• One camp says moving to Rust/Go for JS tooling proves JS can’t viably host its own high‑performance tools (esbuild and the TypeScript-in-Go rewrite cited).
• Another camp pushes back:
  • Languages make domain tradeoffs; not hosting tooling isn’t “failure.”
  • JS is good for fast feedback, UIs, plugins; heavy compilers and linters can live in Rust.
• Broader sentiment: “not everything needs Rust,” but Rust’s safety and excitement make rewrites more likely now than with C/C++.
• Many note this is part of a general trend away from JS-based tooling towards native-core tools that still expose JS plugin APIs.

Performance, architecture, and why now

• Multiple reasons given for “why it took so long”:
  • Need for clean-slate architecture, deep performance knowledge, and real pain from slow tools.
  • Fractured ecosystem and low barrier to entry led to many smaller, slower JS tools.
• Technical points on oxc performance vs SWC/Babel:
  • Arena-allocated AST designed for multiple passes (lint + transform + codegen) with fewer allocations and less GC pressure.
  • Less legacy transform baggage than SWC; data layout tuned from scratch for speed.
  • Anecdotes: big CI pipelines seeing ~8x vs Babel with SWC, plus an extra 30–40% from oxc; claims of extremely fast TS→JS transpilation at large scale (exact conditions debated).

What oxc actually is (and isn’t)

• Clarified repeatedly: oxc is a suite of JS/TS build tools (parser, transformer, TS “type stripper,” linter, formatter, traversal utilities).
• It outputs JavaScript; it is not:
  • A JS/TS runtime (you still need V8/Deno/Bun/browser).
  • A native binary compiler (another project was linked for TS→native via .NET/NativeAOT).

Tooling quality, UX, and overlap with Biome

• Some users report oxc-based tools (especially the formatter) behaving poorly on half-written code and prefer Biome “today.”
• Others note oxfmt is still alpha and ask for patience.
• A complaint: oxfmt with no arguments recursively rewrites all JS/TS files in the current tree. This surprised some:
  • One group says this is expected and documented, and you should always use VCS and/or try tools in a throwaway directory first.
  • Another group calls it bad CLI UX; they expect tool with no args to show --help rather than modify files.
• Question raised why the ecosystem needs multiple Rust-based formatters/linters when Biome already combines linting, formatting, and import sorting.

Broader ecosystem and naming

• Some see oxc as part of a “grand unified compiler” approach: one fast Rust core powering linter, bundler, TS compiler, etc., to avoid duplicated parsers/ASTs.
• Skepticism that Rust-driven tooling speedups coexist with widespread tolerance for slow, bloated browser apps; others argue these are often different people/problems.
• Naming complaints: oxidation/rust/corrosion puns are viewed by some as overused and uninformative, though others note such naming patterns (like py- in Python) are common.