Hacker News, Distilled

Apple Leadership and Corporate Direction

• Some want a hardware-focused successor to the current CEO; others explicitly don’t want leaders associated with aggressive services monetization.
• Several argue Apple’s board, not just design leads, has lost clarity on why Apple is valuable and tolerates missteps.
• There is skepticism that a “new Apple” is emerging; some see the Neo and leadership shuffles as incremental, not transformative.

Design, UX, and Software Quality

• Many feel macOS/iOS UI has regressed: less distinguishable controls, more wasted space, inconsistent idioms, and “Liquid Glass” is a common target.
• Others say day‑to‑day workflows haven’t meaningfully changed; visual complaints are seen as cyclical aesthetics.
• Keyboard text selection, autocorrect, and nagging network-permission prompts are cited as concrete UX degradations.

Authentication & Input Experiences

• Opinions on Face ID and Touch ID are split.
  • Some report Face ID as flawless, even with glasses/facial hair.
  • Others find it unreliable in low light or bright sun and wish Touch ID were still an option.
• Butterfly keyboards and the Touch Bar are widely criticized, but a minority liked them.

Emotional Investment in Apple

• Some compare caring about Apple to sports fandom or national institutions like Bell Labs.
• Others reject corporate tribalism, viewing devices as interchangeable tools and switching brands freely.

MacBook Neo: Purpose and Reception

• Broad agreement it’s a new category for Apple: a genuinely budget Mac notebook, roughly “Apple’s Chromebook.”
• Praised for: low price, good build, strong performance for basic tasks, escape from Windows/Chromebooks, and strong early demand (e.g., preorders selling out).
• Criticisms focus on: fixed 8 GB RAM and future e‑waste, single USB‑C arrangement, lack of Touch ID, and fears of “disposable” mindset.
• Many see it as ideal for students, kids, light office work, and cloud‑centric dev; not for video editing or heavy local workloads.

Steve Lemay and Hopes for Change

• Anecdotes from early macOS days depict Lemay as principled, able to justify design decisions, even if controversial.
• Some fear his long tenure means he shares blame for current design issues; others hope his promotion signals a course correction away from Liquid Glass and recent UI missteps.

Exploit, Firmware, and Practicality

• Linux currently requires a full exploit chain (e.g., Byepervisor) and only works on very old PS5 firmware (around 1.xx–2.xx).
• Latest firmware users likely cannot do this; one commenter notes if a console isn’t on 2.x or below, “he is out of luck.”
• There is mention of a separate userland-only exploit on newer firmware, but it does not provide kernel-level control needed for Linux.

Dual-Boot and Access to Original OS

• People ask whether Linux can coexist with the stock PS5 OS so they can still play their PS5 library.
• The thread does not provide a clear answer or confirmed dual‑boot setup. This remains unclear.

GPU, BC‑250, and Porting Details

• The PS5 uses a custom AMD GPU broadly similar to RDNA2 but reportedly missing some RDNA2 features (e.g., mesh shaders; closer to “RDNA1+RT”).
• A mining board (BC‑250) based on binned PS5 APUs was sold cheaply; prior work on running Linux and AMDGPU/Mesa on this hardware heavily enabled the PS5 port.
• Mesa support for PS5 reportedly hinged on a tiny change (GPU ID range), though people note kernel/platform patches are also needed for PS5’s custom I/O and SuperIO.

Use Cases: Gaming, Media, AI, and Clusters

• Enthusiasm around turning PS5 into a “Steam Machine,” media server, or general Linux box.
• Some ask about running AI models and leveraging the 16 GB shared memory; replies say unified memory alone doesn’t make it a great LLM platform compared to cheap GPUs.
• Historical PS3 supercomputing clusters (e.g., Air Force use) are discussed; opinions differ on whether similar PS5 clusters are likely or practical now.

Locked-Down Hardware, Ownership, and Business Models

• Strong debate over whether it’s “sad” that running your own software on your own hardware is noteworthy.
• One side: consoles are sold at or below cost and subsidized by game licensing, so locking them down is economically rational and expected.
• Other side: PS5 is architecturally a general-purpose computer, and restricting user software is seen as harmful lock‑in and an erosion of computing freedom.

Security, Liability, and DRM Justifications

• Pro‑lockdown arguments: open platforms increase attack surface, malware risk, and reputational damage from users bricking devices.
• Critics argue this is mostly a pretext for profit and control; they note PCs and older phones allowed custom OSes without “the world burning.”
• Some point to liability and certification (e.g., UL) as a partial driver, but others say lock‑down primarily serves vendor business interests.

Hacker Culture and Accessibility

• Many celebrate this as classic “hacker spirit”: making single‑purpose devices do unintended things.
• Others lament decreasing access to open hardware (consoles, IoT, appliances, tractors, 3D printers) and the loss of outlets like big electronics stores.
• There’s disagreement over whether hacker culture has declined or just shifted (more software‑focused, more startup/“build to sell” mindset).

Context: sudden solitude after long-term relationship

• OP: late-30s, remote worker, recently lost a partner of ~20 years, now living alone in the suburbs with a dog, feeling “hollow” and panicky after long stretches without human contact.
• Feels unable to act on standard advice (hobbies, dating, dog park, etc.), despite psychiatric care and multiple psych meds.

Mental health, medication, and grief

• Many see clear signs of depression and grief; emphasize this can persist despite medication and may be “situational.”
• Some warn cocktails of antidepressants/anti-anxiety/mood stabilizers can blunt all emotions or create hollowness; several suggest review or second opinion.
• Others stress basics before or alongside meds: sleep, diet, exercise, sunlight, bloodwork (e.g., vitamins, A1C), and therapy.
• Several note grief comes in waves; acceptance and “working through it” takes months to years and is normal.

Building social contact and structure

• Strong consensus: you cannot stay home and think your way out; you must physically go where people are.
• Suggestions:
  • Gyms, group fitness, CrossFit, martial arts, yoga, running clubs, rock climbing, dance (salsa, tango), improv.
  • Libraries, cafés, coworking spaces, tech meetups, book clubs, board-game nights, language/night classes.
  • Volunteering (food banks, shelters, churches, civic clubs, retirement homes, campaigns).
  • Routine “third places” at consistent times to build familiarity.
  • Consider moving to a more walkable/city environment or renting a room / getting a lodger.

Learning to be alone vs. refusing it

• Split views:
  • Some advocate treating “being alone” as a skill: small intentional solo activities, thin weekend structure, cultivating inner life (reading, journaling, meditation, philosophy, nature).
  • Others argue against becoming too good at solitude, fearing it can lead to permanent isolation and missed relationships.
• Several recommend personal projects and purpose (creative work, self-improvement, passion projects) over passive distractions or games alone.

Micro-connections and coping tools

• Use low-stakes interactions: brief chats with strangers, dog-walk small talk, “lingering” after classes.
• Journaling, gratitude, audiobooks/podcasts/streams, and talking to pets can partially meet the need to “tell someone about my day.”
• Many emphasize self-compassion, patience, and seeing this period as temporary but potentially growthful.

Mac Studio 512GB Removal & Apple’s Strategy

• 512GB RAM Mac Studio was niche and very expensive (~$10K), but some report strong demand in certain channels, especially for AI workloads.
• Main theories for its removal:
  • DRAM price spike makes the config unprofitable or hard to price without PR fallout.
  • Apple sold through its planned production run and doesn’t want to reorder before next-gen chips.
  • Apple may make more money selling multiple 256GB machines instead of one 512GB.
• Some think it’s tied to an impending M5 Ultra Mac Studio (possibly with 512–768GB RAM) and that Apple wants to avoid buyer regret on a soon-to-be-obsolete $10K machine. Exact roadmap is unclear and based on rumors.

Apple Silicon, AI, and Local Models

• High‑RAM Mac Studios are seen as compelling for local LLM inference: unified memory is shared CPU/GPU RAM and cheaper per GB than equivalent GPU VRAM rentals.
• Others argue that for many, slotted RAM PCs with upgradeability or GPU VRAM still make more sense for ML.
• Concern that soldered, non-upgradable RAM forces overbuying “for future‑proofing,” especially painful during a RAM price spike.

RAM Shortage vs. Cartel Debate

• Many attribute rising prices and product changes to AI data center demand for DRAM.
• Others argue the big DRAM vendors act as a de facto cartel via public, coordinated supply cuts, pointing to past price‑fixing cases and long‑term flat real prices.
• A detailed comment claims large AI players pre‑committed a huge share of global RAM supply, pushing up prices and squeezing consumers, open‑source, and small providers.
• There is disagreement whether this is primarily genuine scarcity, coordinated supply management, or both.

Broader Hardware Market Effects

• Raspberry Pi 5 and other SBCs have become much more expensive; several posters now see used x86 mini PCs as better value, with Pis retained mainly for low power, PoE, and ecosystem.
• Some expect RAM shortages to normalize in 1–3 years, as in past cycles; others say this spike is unprecedented and could be “extinction-level” for low-cost electronics makers.

Future Macs & Positioning

• Speculation that Apple will:
  • Release an M5 Ultra Studio (possibly 768GB), and maybe a rethought Mac Pro focused on AI.
  • Continue using high RAM tiers in marketing around local model capability.
• Product naming (Max vs Ultra) and fragmented line (Air/Pro/Mini/Studio/Pro) are seen as confusing.

Title and scope confusion

• Several readers expected notes on hand-writing Wasm, not Rust→Wasm with wasm-bindgen.
• The submission title on HN was edited to add “Rust” for clarity.

Perceived role and adoption of Wasm

• Some argue Wasm has become bloated and niche, drifting from its original “hot loop acceleration” promise toward a general sandbox/component model.
• Others counter that it’s widely used in real products (design tools, CNCF ecosystem, ML in the browser, server-side sandboxing/plugins), so “niche” is overstated.
• There’s disagreement: one side predicts Wasm will remain a small niche; another asks for evidence and notes that Wasm users are often not typical JS developers.

Web interop, DOM access, and JS glue

• A recurring frustration is reliance on JS “shim” code for DOM/Web APIs, especially string conversions (UTF-8 vs UTF-16) and boundary overhead.
• Some want direct DOM and web API access from Wasm to build “traditional” webpages without JS.
• Others argue DOM is inherently slow and string-heavy; JS shims aren’t usually the true bottleneck.

Rust-centric tooling and bindings

• Wasm is seen by some as “basically a Rust thing” today, especially around the Wasm Component Model and WASI components.
• Complaints about wasm-bindgen: complexity, stagnation (with mention of recent renewed work), and bloat compared to nicer experiences like PyO3.
• Multiple comments recommend avoiding automatic high-level bindings (Rust/C++ objects ↔ JS) in favor of C-style interfaces, batched calls, and plain data types.
• Helper tools like tsify are cited as useful for Rust/TS type alignment.

Component Model debate

• Proponents say the Wasm Component Model reduces N² glue complexity by standardizing type interop, improves dev experience, and can make Wasm a first-class web citizen (including potential direct web APIs).
• Critics see it as overengineered (likened to old COM/CORBA), solving a problem OS ABIs already handle with simple numeric types and conventions.
• There’s concern about adding complexity to browsers vs keeping it in toolchains; others reply that standardizing this layer is precisely the point.

GC languages and runtimes

• Discussion notes emerging Wasm GC support and experimental languages, but also current drawbacks: copying overhead across the boundary, poor string handling, and no multithreading yet.
• Some feel GC on Wasm is redundant in browsers (JS already exists) and inferior to mature JVM/CLR GCs outside the browser.
• Others point to existing GC-language targets (Go, Java, C#, Blazor) that mostly ship their own GC in Wasm.

Impact of Claude Code / AI on Teams

• Many expect teams to get smaller; 1 senior engineer plus AI can now do work that previously needed several people.
• Some argue roles (PM, designer, engineer) will be collapsed into “generalist builders” using AI; others insist all roles still matter but fewer headcount will be needed.
• Concern that communication and collaboration may worsen if everyone tries to do everyone else’s job without the old role boundaries.
• Some think AI will “ruin” teams in the short term via over-firing and role confusion; long term impact seen as dependent on leadership’s ability to correct mistakes.

Specialists vs Generalists

• One camp: AI empowers competent generalists; specialists become less necessary, especially below “enterprise-level” complexity.
• Opposing view: sustainable AI use requires strong specialists to control architecture, quality, and complexity; otherwise you get unmaintainable “vibe code.”
• Several note that knowing AI’s capabilities/limits is now a critical skill.

Capabilities and Limits of AI Coding Tools

• Users report large productivity boosts (1.5–5x) for routine feature work, but note review, testing, and integration time remain similar.
• Others say tools are far from “solving coding”; good at generating code, bad at deep design, refactoring, and nuanced problem-solving.
• AI-produced code tends to be more complex than necessary; risk of hitting context-window limits on large, messy codebases.
• Testing by AI is viewed skeptically; real testing is framed as a human investigative activity.

Job Market, Layoffs, and Economics

• Some say layoffs are already happening and attributed to AI-driven cost cutting and “rational reallocation of capital.”
• Others argue broader macro belt-tightening and failing SaaS business models are the main drivers, with AI only accelerating trends.
• Fear that companies will fire engineers, then later face expensive cleanup of AI-generated spaghetti and hire consultants at high rates.

Learning, Skill, and Role Elitism

• Debate over whether programming, design, and PM are universally learnable, or require innate talent.
• Some call programmer exceptionalism “intellectual elitism”; others insist not everyone can reach competence in coding.
• Multiple comments warn that heavy AI reliance may shorten “learning loops,” letting people build faster but actually learn less.

Article framing & prior planning

• Several commenters criticize the article’s title as click‑bait and unsupported (“they all said…it would be brief”), but see the topic as legitimately serious.
• Some note that Hormuz blockade scenarios have been studied for decades and contingency plans/stockpiles exist, so impacts are significant but not civilization‑ending.

Regional economic impacts

• East Asia (Taiwan, Japan, South Korea) seen as most exposed: high dependence on Gulf oil, limited reserves in some nearby countries, and concentration of critical tech manufacturing (TSMC, display fabs).
• Comparisons made to Covid‑scale disruptions: petrochemicals and transport fuel shortages could ripple through global supply chains and trigger financial instability or a credit crunch.
• Some suggest markets under‑priced the risk and may now overreact.

US impact & oil/refining debate

• One side: US is a net petroleum exporter with a partially filled Strategic Petroleum Reserve; direct trade through Hormuz is limited, so strategic risk is modest.
• Counterpoints:
  • US still imports crude and is tied to global prices; domestic consumers will feel price spikes.
  • Disagreement over whether US refineries can efficiently process domestic light sweet crude vs imported heavy sour crude; some argue retooling is feasible, others say infrastructure is optimized for heavy crude.
• Broader concern: real threat to the US is potential Chinese escalation if Chinese oil security becomes existentially threatened.

China, Russia, and energy transition

• Claims that higher oil prices benefit Russia via discounted “black‑market” sales and budget relief, regardless of sanctions.
• Discussion over how much crude China actually imports from Russia vs the Middle East; consensus that China still relies heavily on Gulf oil but has sizeable reserves and electrified logistics (rail, some trucking).
• Hormuz crisis seen as validating China’s push into renewables; EU dependence on fossil fuels vs Chinese critical materials is debated.

Military, shipping, and asymmetric warfare

• Skepticism that naval escorts alone can make Hormuz “safe” enough for insurers and crews, given cheap drones, missiles, and Iran’s geography.
• Concern that degrading Iran’s formal state apparatus could entrench long‑term asymmetric attacks on shipping and possibly aircraft, making the strait permanently hazardous.

Iran war, regime change, and nuclear trajectory

• Many expect the US to eventually declare “victory” without decisive regime change, leaving Iran poorer but intact and more motivated to pursue nuclear weapons, citing North Korea as precedent.
• Others argue Iran will be continually “contained” by periodic strikes and won’t be allowed to go nuclear.
• Thread notes deep mistrust: US and Israel are accused of negotiating in bad faith and striking during talks; others claim Iran also drags out negotiations.
• Reported statements from Iranian officials and mediators (e.g., Oman) are cited to argue a recent near‑deal was derailed by US attacks.
• Fears raised that eliminating any prospective leadership and threatening future leaders creates a governance vacuum for ~90 million people and locks in endless guerrilla conflict.

Israel, US politics, and regional alignment

• Some predict long‑term erosion of unconditional US political support for Israel, especially among younger conservatives, though change is expected to be slow.
• Others justify hard containment of Iran as necessary given Iran’s longstanding hostility toward Israel, while critics question why Western publics should privilege Israel’s security over other regional states.
• Middle Eastern governments are portrayed as short‑sighted for not coordinating more effectively to check US‑backed Israeli actions, with speculation they will regret this later.

Energy transition, nuclear, and climate

• Hormuz crisis prompts renewed arguments for:
  • East Asian countries (notably Taiwan and Japan) to expand nuclear power instead of relying on seaborne fuel imports.
  • Accelerating renewables and electrification to reduce strategic reliance on oil chokepoints.
• Debate over centralized nuclear vs decentralized solar/wind in wartime; transformers and grids remain vulnerable either way.
• Some hope that forced reduction of oil use could reduce “dumb wars” over hydrocarbons, while others stress coal‑to‑liquids and petrochemical pathways as fallback strategies, especially for China.

Overall sentiment

• Mix of alarm and dismissal:
  • Alarmed commenters foresee systemic supply shocks, destabilized financial markets, and a durable rise in asymmetric warfare around Hormuz.
  • Skeptics emphasize preparedness, alternative supplies, and adaptability of refineries and logistics, seeing the crisis as serious but manageable rather than globally catastrophic.

Maintenance status of PyPy

• Original concern: a tool (uv) labels PyPy as “unmaintained,” which some see as overstating the situation.
• Several commenters distinguish:
  • “Unmaintained” / “dead” vs.
  • “Low activity” / “not under active development.”
• Data cited: a few commits per month since late 2025 and a release in July 2025; some see this as reasonable for a volunteer project.
• A core developer states PyPy is maintained (bugs fixed, JIT occasionally improved) but current core devs cannot keep up with new CPython versions without more contributors.

Version lag and ecosystem compatibility

• PyPy targets Python 3.11 while CPython is at 3.12+; being multiple minor versions behind is seen as risky for long‑term viability.
• Some argue PyPy has always lagged behind and that this isn’t new.
• Others counter that the lag now pushes PyPy out of many libraries’ official support windows, especially in scientific Python, making it effectively unusable for much of the 2026 ecosystem.
• There is debate over whether slow progress toward 3.12 support means the project is “not actively developed” vs. simply under‑resourced.

Technical value and limitations

• Many praise PyPy’s speed (often several‑fold faster on CPU‑bound pure‑Python code) and its research contributions (meta‑tracing JIT, RPython, influence on HPy and CPython internals).
• Users note it works well for pure‑Python code, but there are major pain points with C extensions built on the CPython C API (NumPy, SciPy stack, crypto libs), where PyPy’s cpyext layer is slow or incomplete.
• Some describe serious incompatibilities (e.g., garbage collection behavior causing resource exhaustion) and poor documentation of such differences, calling PyPy “basically useless” for large apps needing full CPython interoperability.

Funding, sustainability, and contributions

• Multiple comments lament chronic underfunding and lack of corporate support despite widespread indirect use of Python.
• Suggestions include more prominent donation links, sponsorship tiers, and clearer calls for corporate funding or hiring dedicated maintainers.
• A core dev invites both financial and code contributions and notes active work toward Python 3.12 with new contributors.

Naming confusion and side topics

• Frequent confusion between PyPy (interpreter) and PyPI (package index), with some arguing the similar names are a persistent usability issue.
• Brief side discussions on AI tools for OSS maintenance and whether alternative interpreters (e.g., RustPython) are better bets going forward.

Is Sci-Fi Dying?

• Many disagree with the article’s “death” framing; they see a genre in transition, not collapse.
• Some argue that reality now resembles old sci‑fi (AI, climate, billionaires, geopolitics), which both blunts the sense of wonder and creates new angles to explore.
• Others feel there is already a lifetime of unread “back catalog” and are unconcerned about new output.

AI, Authorship, and Economics

• AI is seen as both a topic and a tool: fertile ground for stories, yet also a threat via content floods and automation.
• Several say AI-generated fiction feels programmatic and hollow; they want human-authored prose, even if AI assists planning.
• Economic concern: even “perfect” AI books would drown in oversupply, undercutting discoverability and income.

Politics, Sexuality, and “Lecturing”

• Long-running tension: some readers are tired of overt politics/sexuality and didactic tones.
• Others counter that these themes have always been central to sci‑fi; what changed is which politics/sexualities are foregrounded.
• The dividing line for many is not topics but heavy-handed moralizing versus subtle, reader-respecting treatment.

Utopias, Dystopias, and Tone

• Several lament the dominance of dystopia and crave optimistic, “noblebright” adventure and functional utopias.
• Others argue pure utopias are hard to dramatize; conflict-free worlds feel dull.
• There’s a sense that “dark and edgy” has become the easy way to appear profound, and a pendulum swing back to hopeful stories is anticipated.

Quality, Tropes, and Awards

• Many see good sci‑fi as rare amid a sea of tropey, shallow, power-fantasy pulp—but still worth sifting for the standouts.
• Some distrust major awards and imprints, viewing them as politicized or commercially constrained, and look instead to overlooked or outsider works.

Medium Shift and Discovery

• Commenters highlight strong contemporary sci‑fi in TV/streaming and especially video games, which may be the most vibrant current medium.
• Indie platforms, web serials, and fanfiction communities are cited as thriving spaces for experimental or more positive sci‑fi.

AMD CPUs and AI positioning

• Commenters praise AMD’s EPYC line: Genoa was a big jump over Milan, and Turin is seen as an even larger leap, with very strong performance in cloud VMs and on-prem.
• Some criticize AMD for “dropping the ball” on AI accelerators and software compared to NVIDIA’s ecosystem, arguing that software stack matters more than hardware specs.
• Others counter that AMD does produce competitive AI chips and is landing major partnerships, suggesting the AI story isn’t as dire.
• Several participants say they “don’t care” about accelerators as much as they care that AMD broke Intel’s CPU dominance and brought core counts and prices back to sanity.

---

Cloud vs self‑host / colocation economics

• Many argue big cloud (especially AWS/GCP) is dramatically more expensive than owning/racking hardware or using dedicated servers/colo, particularly for steady workloads like CI or databases.
• Counterarguments emphasize total cost of ownership: datacenter space, power, cooling, hardware failures, hands-on ops, provisioning delays, redundancy, and staff time.
• Cloud is seen as paying for instant provisioning, autoscaling, multi-region, managed services, and powerful APIs; it shines for spiky/unpredictable load and fast iteration.
• Opinions differ on when self-host becomes cheaper: some cite $1–3M/month cloud spend as the tipping point; others report big savings even at modest scale by moving to colo or smaller providers.
• Hybrid approaches are suggested: run base load on owned/dedicated hardware and burst to cloud for peaks.

---

Oracle Cloud: cheap but distrusted

• Several are wary due to Oracle’s reputation for aggressive licensing and lock‑in; one calls Oracle broadly “predatory.”
• Some still use Oracle Cloud for very small, portable projects because compute can be extremely cheap, as long as no proprietary services (especially Oracle DBaaS) are used.
• Reports of poor UX: confusing signup, free tier instances reclaimed for “idleness,” trials abruptly shut down, and inconsistent account handling.
• A large migration case (hundreds of VMs) reports ~40% cost savings vs AWS/GCP by using only basic primitives (compute, storage, load balancers), free egress, flexible CPU/RAM combos, and flat regional pricing—but warns that many higher-level managed services are unreliable and the platform UX is “nightmarish.”

---

Hetzner, OVH, and other non‑hyperscale options

• Hetzner is repeatedly highlighted as having outstanding performance per dollar, especially for dedicated servers; many say a cheap Hetzner box would top the benchmark charts.
• Caveats: more initial work (no-frills bare metal, you handle DR, hardware failures, encryption), slower provisioning than cloud VMs, and at least one warning not to use Hetzner for “anything actually important” due to past incidents.
• OVH is frequently mentioned as a similarly priced or cheaper alternative with strong dedicated offerings; experiences with support are mixed but some report very reliable use in production and testing.
• Other providers surfaced include Vultr, HostHatch, netcup, and various “low-end” hosts; opinion on Vultr is split between “great for years” and “expensive with annoying limits.”
• Several tools and comparison sites exist to navigate fragmented pricing, benchmark low-end hosts (e.g., via common scripts), and automate cross‑cloud/right‑sizing choices.

---

Benchmarking nuances and missing dimensions

• Multiple comments stress that VM benchmarks measure not just CPU but hypervisor behavior, CPU pinning, storage backend, network virtualization, and noisy neighbors; these can shift results 20–30%.
• vCPUs often don’t match bare-metal CPU capabilities or exposed feature sets; cloud providers sometimes mask features to enable live migration.
• Some note that consumer/gaming CPUs (e.g., high‑end Ryzen/desktop chips) can significantly outperform many cloud VMs in single-thread and memory performance.
• Participants want additional benchmarks: network throughput and egress pricing, SAN/non‑local storage performance, GPU training/inference workloads, and AWS burstable instance families (t4g vs competitors).

---

Debates on writing style and AI-generated content

• A side thread critiques a popular “we moved from AWS to Hetzner” article as likely AI‑generated “slop,” citing repetitive marketing phrasing.
• Others argue repetition is a normal rhetorical device and that dismissing content solely for being AI‑assisted is unhelpful; accuracy and effort should matter more.
• Several express frustration that AI‑style, overlong, promotional writing wastes readers’ time and erodes diversity of style, even when technically correct.

Progress, Automation, and “Meaningful Work”

• Some argue technological progress is political, not inevitable, and often concentrates wealth while increasing unemployment.
• Others counter that most people from the past would choose today’s comforts, medicine, and non-manual work; “romanticizing” physical labor ignores how damaging many manual jobs are.
• A minority suggests “less tech” could itself be progress, by improving resilience and meaningful physical activity, but this is strongly challenged.

Unions, Power, and Worker Leverage

• Several posters say software workers should have unionized earlier; now AI erodes leverage.
• Skeptics argue unions can’t help when the labor itself is no longer needed.
• Some think unions could still improve layoff processes and severance, even if they can’t stop automation.

What If Software Jobs Shrink or Disappear?

• People consider fallback careers: psychology, accounting, trades (electrician, welding, nursing, landscape design), piloting, hobby farming, small retail/food businesses.
• Entrepreneurship and “lifestyle businesses” are frequently mentioned; AI is seen as a force-multiplier for solo founders.
• There is fear of personal ruin, especially around housing costs; others reply that housing prices would adjust if high-paying jobs vanish.

How Far Can AI Go? Capabilities and Limits

• One camp sees LLMs already better than an “average” developer in many tasks, predicting large job losses even if they never reach perfection.
• Another camp stresses hallucinations, brittle reasoning, lack of real-time learning, and guidance overhead; they doubt full replacement and expect persistent demand for quality human work.
• Debate over “just a text predictor”: some say that’s fundamentally limiting; others note that prediction-plus-feedback is already surprisingly powerful.

Future Shape of Software Work

• Many expect fewer developers overall, with remaining roles more like architects: deciding what to build, orchestrating AI, navigating organizations, validating solutions.
• Concern that if routine coding is automated, juniors lose the traditional ladder to senior/architect roles.
• Some think SWE may be hit first but other white‑collar domains (accounting, office work, basic IT, marketing) will follow.

Broader Economic and Social Impacts

• If most office jobs are automated, posters foresee severe knock-on effects: commercial real estate, local services, and small businesses hollowed out.
• There is skepticism the US will deliver generous safety nets or UBI; instead, several advocate aggressively enabling AI-powered small-business formation to avoid social crisis.

Emotional Reactions and Values

• Thread mixes anxiety, resignation, and cautious optimism.
• Some accept that if a job can be automated it “should” be, but mourn loss of joy in problem-solving.
• Others are optimistic about 10x productivity, more solo projects, and a possible cultural shift back toward people in tech who “love the game.”

Capabilities and Use Cases

• Many users find Qwen3.5 surprisingly strong for local coding: multi-file edits, building small Rust apps, HTML/CSS work, and “rubber duck” explanations of code and compile errors.
• 9B and 4B/0.8B variants are used for OCR, text cleanup, simple coding, translation, and log triage, but are considered weak for complex coding or “agentic” multi-step tasks.
• Qwen models are also used for structured text/image/video analysis (JSON output, NER, categorization) and financial data extraction from emails.
• Several note that Qwen3.5 is very confident and often sycophantic; personas (e.g., “brief rude senior”, “emotionless vulcan”) help tone it down.

Performance, Hardware, and Context

• Reports span from tiny SBCs and phones (0.8B/2B CPU-only) to RTX 30/40/50-series, Apple M1–M5, and A100/H100.
• 9B often reaches ~60–100 tok/s on mid/high-end GPUs; 35B-A3B can run at ~14–25 tok/s on consumer cards with partial offload.
• 27B and 35B-A3B are widely seen as the sweet spot for “serious” local coding, with 27B sometimes stronger on benchmarks, 35B-A3B faster due to MoE.
• Long contexts (100k–256k) are possible but some see quality degradation and instruction drift over long sessions, partly due to sliding-window attention.

Quantization and Model Choice

• Frequent discussion of Q4_K_M / UD-Q4_K_XL vs other 4-bit schemes; Q4_0 and Q4_1 are described as faster but notably less accurate.
• Rule of thumb in the thread: more parameters at lower bit-depth usually beats fewer parameters at higher bit-depth, down to ~3–4 bits.
• Users recommend experimenting: 27B at 4-bit or 35B-A3B at 3-bit for 16GB+ VRAM; 9B or 4B for very constrained setups.

Thinking Mode and Reliability

• “Thinking” mode can run indefinitely or add large latency; multiple users disable or heavily constrain it.
• Some note improved reasoning and code review with thinking enabled; others find loops, repetition, or crashes in certain orchestrators.

Comparison to Frontier and Overall Sentiment

• Enthusiasts claim 27B/35B approximate older Claude/GPT tiers or at least Haiku-level for many coding tasks, and can save substantial API costs.
• Skeptics report that even 122B/397B lag behind top proprietary models (e.g., Claude Opus/Sonnet) on hard coding and mathematical/physical reasoning, and still hallucinate.
• Consensus: Qwen3.5 is a major step for local models, excellent for many targeted workflows, but not yet a full replacement for state-of-the-art hosted models—especially for complex, long-horizon “agentic” coding.

Core proposal: ZIP/postcode first

• Many agree the basic idea—using a high‑information field early to auto-fill city/state/country—can reduce typing and feel snappier.
• Several note this is already common in some countries (e.g., UK, NL, IE, JP) where postcodes are fine-grained and backed by good postal databases.
• Some suggest a refined pattern: country first (possibly prefilled via IP), then postal code, then the rest.

Limitations of ZIP codes (even in the US)

• Multiple comments stress: ZIPs are about delivery routes, not political boundaries.
• ZIPs can:
  • Map to multiple cities and even multiple states.
  • Change over time.
  • Cover PO boxes only or unusual entities (boats, large campuses).
• USPS treats each ZIP as having a “preferred” city and alternates; mail works even with “wrong” city, but legal/municipal boundaries often differ.

International & format issues

• Many point out that:
  • “ZIP code” is US‑specific; other countries use postal codes with different formats (length, letters, digits, or none at all).
  • Numeric-only inputs and 5‑digit assumptions break for Canada, much of Europe, etc.
  • The claim that 5 digits determine the country is flatly rejected; the same codes exist in multiple countries.
• Non‑US readers often see US‑centric design as actively hostile or unusable.

UX, autofill, and validation

• Several argue built‑in browser autofill is simpler and often superior: no typing at all if forms are standard.
• Over‑eager validation and “corrections” frequently reject valid foreign addresses or diacritics, causing drop‑offs.
• Auto-fill should be suggestion, not hard validation; users must be able to override guessed city/state.

Data & implementation complexity

• Maintaining accurate global postcode→place mappings is non‑trivial; commercial datasets are incomplete, expensive, and change.
• For US‑only or single‑country sites, ZIP/postcode‑first can work well if backed by postal data and still editable.

Alternative approaches

• Some favor:
  • Country → postcode → structured fields (with soft autocomplete).
  • Or even a single free‑text, multi‑line address box parsed server‑side.
• Others note address entry isn’t a huge burden; “clever” flows often fail more than they help.

Tone and reception

• Many criticize the original site’s US‑centric framing and later snarky geo‑targeted messages as condescending.
• Nonetheless, several commenters see merit in the underlying principle if adapted carefully and globally.

Resignation and Stated Reasons

• Resignation framed as driven by principle: concern over AI in national security, specifically:
  • Mass surveillance of Americans without judicial oversight.
  • Lethal autonomous systems operating without human authorization.
• Some commenters respect the decision and see it as a rare example of acting on principle.
• Others see it as selective or late, noting:
  • OpenAI already had military ties earlier.
  • The person previously worked at other large tech firms with similar issues.
  • Suspicion they joined post-ChatGPT mainly for money/RSUs and are now reputationally repositioning.

Principles vs People

• Debate over the claim that this is “about principle, not people”:
  • Some argue you can strongly disagree with actions yet still respect colleagues and be open to working with them if they change.
  • Others say trying to take a strong ethical stand while keeping relationships intact is inconsistent or self-serving.

Ethics of AI, Warfare, and Surveillance

• One camp: Any work at OpenAI (and similar firms) now contributes to:
  • Autonomous weapons.
  • Mass surveillance.
  • Social-control infrastructures (e.g., “social credit”–style systems).
• Counterpoint: Advanced, precise AI-enabled weapons and systems could:
  • Reduce collateral damage compared to older, cruder methods (e.g., carpet bombing).
  • Save soldiers’ lives by automating dangerous roles.
• Strong pushback: history of drones shows “low-cost” war increases intervention, not restraint.
• Dispute over red lines:
  • Some see any “death tech” as immoral, with or without a human in the loop.
  • Others focus specifically on domestic surveillance and “killbots” as special dangers to democracy.

Double Standards and Scope of Responsibility

• Questions raised:
  • Is objecting only to spying on Americans a coherent moral stance vs spying on foreigners?
  • Should responsibility extend equally to mass surveillance abroad?
• Comparisons to other tech companies (Anthropic, Google, Microsoft, Palantir):
  • Argument that similar critiques should apply broadly, not only to OpenAI.
  • Counterargument that working on general-purpose products vs core AI models for defense is not morally equivalent.

Jobs, Morality, and Complicity

• One view: Anyone who stays at OpenAI now is complicit in building tools for killing and control, especially given their high pay and employability.
• Opposing view: People have mortgages, families, and career constraints; moral purity is a luxury.
• Counter-counter: OpenAI-level engineers are highly employable, so “no choice” arguments ring hollow.
• Broader debate about whether morals actually influence people’s choices in a capitalist system, or mostly serve as post-hoc rationalizations.

Geopolitics and China–US Comparisons

• Some argue AI militarization is inevitable because of great-power competition (especially with China); better “our” side build it first.
• Others reject this arms-race logic as the same thinking that fueled WWI and the nuclear arms race.
• Intense back-and-forth on which country (US vs China) is more dangerous or morally worse, with:
  • References to wars, alleged genocides, surveillance states, and historical interventions.
  • No clear consensus; several commenters call both systems deeply problematic.

Social Media and Communication Style

• Some criticize continued use of X/Twitter by people taking moral stands, calling it inconsistent; others see no equivalence between leaving a job and leaving a platform.
• Multiple commenters claim the resignation tweet “sounds like AI-written text,” citing stylistic tropes.
• Broader concern that people increasingly offload even short, personal statements to AI.

OpenAI Governance and Nonprofit Structure

• Skepticism about the original nonprofit oversight structure:
  • Perception that it failed or was co-opted.
  • Comment that the board “did try,” but was replaced, seen as ironic given its mission to control powerful AI.
• Noted that the story reached mainstream news in at least one European country, showing its broader impact.

Meta and Cultural Framing

• Some see this as the new “Why I left Google” genre for the AI era.
• Others express fatigue and cynicism about public resignation narratives framed as high principle while still carefully preserving career options.

Perceived LLM Writing Tropes

• Commenters recognize recurring constructions: “It’s not X — it’s Y,” car‑ad / movie‑trailer tone, faux profundity, overuse of words like “genuine,” “honestly,” “quietly,” and metaphor-heavy phrasing (“tapestry,” “camaraderie”).
• Structural tells include: title–subtitle headlines with colons, parenthetical “(why this matters)” endings, overlong README files, pros/cons tables for trivial points, “The [Noun] [Noun]” headers, “This changes everything,” and heavy first/second person (“We’ve all been there,” “Your first instinct…”).
• Different models have distinct quirks (e.g., “I’ll shoot straight with you,” “Fair enough,” snarky personalization, tech metaphors for everything), but converge on similar persuasive, flattened styles.

Why Tropes Arise (Speculated in Thread)

• Several participants blame instruction tuning / RLHF and mode collapse: base models are said to be more stylistically varied, whereas post‑training pushes toward a single “highly rated” style.
• Human raters and chat-oriented training may reward dense, “helpful” language, leading to nominalizations, longer words, melodrama, and repeated rhetorical tricks.
• Some suggest polluted training data (SEO slop, AI‑generated text, corporate/press‑release style) and lack of diverse incentives.

Detection, Overdiagnosis, and Limits

• Many think raw LLM output is easy to spot today due to bad/flat writing and trope overuse, but expect this to get harder.
• Others warn of a self-reinforcing “AI witch hunt”: people call any polished or em‑dash‑using text “AI,” with no way to verify authorship.
• There’s concern about the information ecosystem being flooded with mediocre auto-generated content, but some argue that many human posts are already low value.

User Strategies and Prompting Challenges

• Negative instructions (“don’t do X”) often fail or backfire; models seem to fixate on mentioned tropes (“pink elephant” effect).
• Suggested mitigations: treat LLM output as a first draft, then manually edit or use a separate “editor agent” to strip tropes; specify a human writing style instead of lists of bans.
• Some users adjust model “personality” settings (e.g., less warmth/enthusiasm) or self-host media and tools to avoid AI‑slop and engagement-optimized platforms.

Attitudes Toward AI-Assisted Writing

• One camp sees AI-written prose as aesthetically bad, manipulative, or even deceitful if presented as personal writing.
• Another camp prioritizes clarity and usefulness over provenance, viewing LLMs as acceptable collaborators or superior to much human output, while acknowledging broader societal and incentive problems.

Scope and intent of the ban

• Bill targets the President, VP, Members of Congress, and senior officials, barring them from trading event contracts.
• Supporters see it as analogous to bans on athletes betting on their own games, to avoid “match fixing” in governance.
• Some note it also prohibits betting on events officials personally participate in.

Corruption and incentive risks

• Strong concern that officials could:
  • Trade on non-public information (election, war, sanctions, regulation, etc.).
  • Directly influence outcomes they have bet on, including catastrophic ones (e.g., war, “assassination market”–style incentives).
• Commenters point to existing political insider trading and say prediction markets are just a more transparent manifestation of ongoing corruption.
• Others argue that even transparent knowledge of corrupt bets doesn’t prevent harmful decisions once money is on the line.

Who should be covered

• Many argue limiting the ban to elected/senior officials is insufficient:
  • Appointees, career bureaucrats, military/intelligence personnel, and even low-ranking staff can hold valuable inside information.
  • Relatives, proxies, “second cousins,” or sham identities could be used to route around any ban.

Transparency vs prohibition

• One camp favors full transparency:
  • All bets tied to real identities, visible in real time, with AML-style rules against obfuscation and harsh penalties for fronting.
  • Idea: journalists and the public could see if officials pile into a market, partly neutralizing insider advantage.
• Critics respond:
  • Proxies and fake identities are easy for powerful actors.
  • Public trade data could expose people to targeting and extortion.
  • Transparency alone doesn’t remove incentives to “throw the match.”

Value and future of prediction markets

• Fans say markets can aggregate dispersed information, improve forecasts, inform personal and business decisions, and act as hedges.
• Skeptics see them as:
  • Mostly gambling with thin liquidity outside a few areas.
  • Intrinsically corrupting, accentuating “gamble on everything” culture and upward wealth transfer.
  • Likely to become niche or discredited as insiders and manipulators dominate.

Enforcement, realism, and broader context

• Doubts that this will matter while stock-trading by officials remains largely allowed and lightly enforced.
• Some argue the core problem is failure to enforce existing fraud/insider laws and broader “money in politics,” not specific new bans.
• Others think it’s easier to restrict prediction markets now, before they become entrenched, and see value in incremental reform despite cynicism.

Incentives, Leaks, and National Security

• Many argue prediction markets create unusually strong financial incentives to leak classified or sensitive information, especially about war and foreign policy.
• Using insider national-security information to bet is seen by some as equivalent to leaking, and should be punished as such.
• Others counter that “markets aren’t the problem”; banks and stock markets already incentivize crime, and the solution should be enforcement and penalties, not banning markets.
• A key concern: there’s no clean way to separate “good” leaks of corporate info from dangerous leaks of government or military plans.

Comparison to Other Financial Markets

• Some say this is overblown because war risk is already traded implicitly via oil futures and other derivatives; prediction markets just make it explicit.
• Counterpoints:
  • Prediction markets give a more direct “if I do X, I profit” incentive than broad instruments like oil futures.
  • Commodity futures are argued to have real hedging value, while prediction markets are often described as thinly veiled gambling.
• There’s debate over derivatives in general: some see them as socially useful and ancient in origin; others view them as legalized gambling that worsens inequality.

Evidence of Insider Trading in Iran-Related Markets

• The thread heavily debates the Polymarket contracts tied to a U.S. strike on Iran and the death/removal of Iran’s supreme leader.
• Some users checked order books and historical prices and claim:
  • Pre-attack probabilities never got very high (e.g., under ~30%).
  • Activity around notable trades did not stand out in context.
  • It was widely predictable from public signals: troop and carrier movements, political “red lines,” and prior Venezuela precedent.
• Others note concentrated large bets shortly before events and see this as circumstantial evidence that insiders may have acted.
• There is disagreement over whether the specific example chosen in the article is strong or weak evidence of insider abuse.

Gambling, Corruption, and Social Harm

• Multiple commenters see prediction markets as part of a broader gambling explosion (sports betting, meme stocks, loot boxes), exploiting addiction rather than improving forecasting.
• They worry markets can reward socially harmful behavior: incentivizing leaks, corruption, even assassination attempts.
• Some highlight Kalshi’s refusal to pay out on a death-related event as a narrow attempt to avoid incentivizing killing, but consider the overall concept of real-world betting “ludicrous” and socially negative.

Regulation, Policy, and Future

• Comments note weakened regulation (e.g., CFTC) and selective enforcement, making abuse more likely.
• Ideas raised include: banning officials from using prediction markets, stricter penalties for insider use, or outright banning such platforms.
• Others suggest states will themselves try to manipulate or exploit these markets, turning them into tools of information warfare.

Scope of the problem: AI, essays, and grading

• Many see essay-writing as already distorted by formulaic rubrics (e.g., SAT-style 5‑paragraph essays) that reward structure over insight.
• Others argue that formulaic writing is still a useful communication skill, especially for argument and analysis.
• Several commenters think traditional take‑home essays are now “non-viable” as assessments due to AI; others insist essays remain viable if done in person with unseen prompts.

AI detectors and false positives

• Multiple anecdotes of purely human writing being flagged as AI, sometimes with the detector highlighting the only genuinely human line.
• Specific quirks: detectors over-weight “big words,” em dashes, or HN/Reddit-like style, making competent or distinctive prose look “AI-like.”
• Concern that non-native speakers or neurodivergent students may be disproportionately misflagged (cited from linked summaries of external reporting).
• One side claims state-of-the-art detectors work reasonably well in a probabilistic sense; the other side counters that any nontrivial false-positive rate is unacceptable for punishment, likening it to fortune-telling or predictive policing.
• Profit incentives for vendors and black-box behavior are criticized; the cost of errors falls on students, not companies.

Behavioral and social effects

• Some people deliberately “write worse” or simplify online to avoid accusations of being AI.
• Recurrent theme: once someone yells “witch,” it’s nearly impossible to disprove; AI accusations are seen as a new witch-hunt.
• Commenters worry students are being normalized to automated surveillance and opaque systems making high-stakes judgments.

Rethinking assessment

• Proposals:
  • Move grading to proctored in-person work: timed essays, oral defenses, monitored writing sessions, or code modification exams.
  • Use take‑home work as ungraded practice/prep; rely on exams for grades.
  • Increase frequency of low‑stakes in-class tests instead of one high‑stakes final.
• Counterpoints:
  • Single end-of-term exams are seen as unfair to students with bad days, health issues, or weak time management.
  • Continuous assessment with feedback is argued to improve learning but is now more vulnerable to AI “expediency.”
  • Some suggest focusing less on “catching cheaters” and more on designing tasks requiring genuine understanding, originality, and personal engagement.

Purpose of writing in education

• Strong view that the act of writing—organizing thoughts, practicing argument—is the core skill; outsourcing to AI undercuts learning.
• Others note that well-used AI may be acceptable if a student can explain and defend the result; if teachers can’t tell, perhaps the assignment needs redesigning.

Overall role and impact of Docker

• Seen as a major cultural and operational shift: “ship your machine” made deployment faster and bypassed traditional ops bottlenecks.
• Some argue Docker simply encoded pre-existing tribal deployment steps in a repeatable form (Dockerfile) rather than making deployment intrinsically “harder” or “easier.”
• Others view it as a hacky but effective workaround for a broken Linux userspace and dependency model.

Containers vs traditional packaging / OS design

• Debate over whether containers improve efficiency: they share the host kernel and can be lighter than VMs, but per-app userspaces duplicate libraries and tools.
• Critics say this is bloat and that static linking or bundling deps beside binaries (as on Windows) is conceptually cleaner.
• Defenders argue isolation and version pinning justify the duplication, especially in multi-tenant or fast-moving environments.
• Some prefer traditional distro packaging + systemd units for single-tenant or small-scale setups.

Nix, Guix, and alternative paradigms

• Nix/Guix praised for hermetic, reproducible builds, avoiding dependency conflicts and enabling fine-grained sharing.
• Others note Nix has a learning curve, documentation gaps, and struggles with complex ecosystems like Python.
• Discussion of more radical alternatives (Plan9/Inferno, unikernels) as ways to “fix the stack” instead of wrapping it.

Dockerfiles, build tooling, and reproducibility

• Dockerfile’s flexibility and shell-based model seen as both its strength and its source of non-determinism and bad practices.
• Some wish for declarative, language-neutral build tools; others point out those historically failed to gain adoption.
• BuildKit, LLB, and third-party tools are used to improve caching, reproducibility, and layering; reproducible container builds remain nontrivial.

Performance, bloat, and ML workloads

• Image sizes, especially for ML (e.g., multi-GB Torch/TensorFlow stacks), are a growing concern.
• People experiment with distroless bases, hardened/minimal images, deduplicating registries, and Nix-based layering to tame size and startup time.

Networking, platforms, and tooling quirks

• Historical use of SLIRP/VPN-like tricks to bypass corporate firewalls is widely discussed and admired as a clever hack.
• Mac container networking and separate IPs remain awkward, often solved with WireGuard, third-party tools, or Tailscale.
• Docker’s manipulation of iptables and ignoring host firewalls (e.g., ufw) is criticized as dangerous.

Thermal and Frequency Limits

• Several comments explain why ~5–6 GHz seems like a wall: dynamic power scales with capacitance × voltage² × frequency, so higher clocks explode power and heat.
• Overclocking records near 9 GHz use liquid nitrogen/helium and hundreds of watts per core, seen as irrelevant to reliable consumer chips.
• Some argue 10 GHz CPUs “will never be done in silicon”; others think it would require exotic cooling, new semiconductors, optics, or reversible/cryogenic computing.
• Interconnect delay is a hard constraint: at 10 GHz a cycle is ~0.1 ns, only a few cm of signal travel.
• Newer process nodes reduce capacitance and can shift the “sweet spot” frequency, but gains in clock rate have been tiny since ~2003.

Single-Core Speed vs Parallelism

• Debate over whether 10 GHz is even desirable for everyday workloads.
• One side: modern tasks are multi-threaded; adding cores and SIMD (AVX, GPUs) gives better returns than chasing clocks.
• Counterpoint: “single-thread” or, more precisely, dependent-operation performance still matters; only higher frequency really helps long dependency chains.
• SIMD/vector/matrix units are called out as parallelism, not true single-thread latency improvement; they shine only with many independent operations.

Other Performance Breakthroughs

• Many say the HDD→SSD transition was a bigger real-world leap than any CPU clock bump after the MHz wars.
• Dedicated GPUs and later high core-count CPUs (Threadripper, etc.) gave huge speedups for specific workloads like games, compiling, and data processing.
• Apple’s M1/M2 laptops are often cited as the first “wow” upgrade since SSDs, mainly for power efficiency and quietness, though not matching top desktop GPUs.

Historical Context: Athlon and Intel

• The 1 GHz milestone is seen as mostly marketing; its real importance was AMD beating Intel and offering better IPC (especially in floating point).
• The Pentium 4’s high clocks are described as “cheating” via very long pipelines and low IPC; Intel originally talked about 10 GHz but hit a wall.
• Intel later retreated to Pentium III–derived cores (Pentium M → Core → modern Core) and adopted AMD’s x86‑64, effectively abandoning the NetBurst/Itanium vision.

Software Bloat and User Experience

• Strong sentiment that software (e.g., Electron/JavaScript apps, chat clients) has absorbed decades of hardware gains, sometimes feeling slower than lean 90s equivalents.
• Others distinguish genuine new capabilities (rich media, DAWs, modern games) from pure bloat.
• Overall sense: raw compute has grown massively, but everyday responsiveness often depends more on storage, OS overhead, and software efficiency than GHz.