Hacker News, Distilled

Framing the CEO’s Limits: Hypocrisy vs Normal Parenting

• Many see “YouTube CEO limits kids’ social media” as obvious: all good parents limit harmful or addictive stuff (compared to soda, candy, cigarettes, alcohol, x‑rays).
• Others argue there is a story: a chief of an engagement-maximizing product publicly acknowledges it must be limited for his own kids, contradicting the “harmless fun / educational” marketing, especially for kids.
• Some emphasize this isn’t total bans: both current and former YouTube leaders reportedly use time limits or kids’ modes, i.e., “everything in moderation.”

Harms of Screens & Social Media (Especially for Young Kids)

• Multiple parents describe iPads and YouTube for young kids as “normalized neglect,” some even call it “abuse.”
• Reported harms: expectation of constant stimulation, stunted emotional development, fine-motor and executive-function issues, tantrums when screens removed.
• Short-form video and algorithmic feeds are seen as especially “brainrotting,” often likened to cigarettes; others say “brainrot” more narrowly refers to low-effort content.
• Several distinguish between:
  • screens for young kids,
  • short-form feeds for teens, and
  • older-style peer-group social media, arguing impacts differ.

Parents’ Responsibility vs Systemic and Economic Factors

• One side: “Just parent harder” – set boundaries, use parental controls, ban or whitelist content, fill time with sports, hobbies, and imaginative play.
• Counterpoint: this underestimates exhaustion, lack of knowledge, and the power of products engineered to be addictive; many parents are caught in the same attention traps.
• Inequality angle: wealthy families can buy childcare, therapy, and tech literacy; poorer kids may be most exposed and least protected.

Tools, Tactics, and Workarounds

• Strategies mentioned: strict time limits; no smartphones for young kids; banning certain platforms (e.g., Roblox); using Switch instead of phones; Plex/local mirrors of approved videos; YouTube Kids with whitelist mode; Apple/Google parental controls.
• Some find these tools powerful; others describe them as confusing, easy for kids to bypass, and requiring constant vigilance.

Peers, Culture, and Regulation

• Peer pressure is a major problem: kids risk social exclusion if they’re off the dominant apps/games.
• Some advocate treating social media more like regulated vices; others fear this becomes a pretext for censorship and state control.
• Several conclude that no law or tech can substitute for active, present parenting.

Scope and severity of the lockout

• Commenters see this as a particularly bad case: decades of purchases, photos, devices and a developer account effectively disabled.
• Many stress the distinction between “closing an account” and “confiscating access to data and devices”; several compare it to a bank seizing deposits.
• The inability to get a concrete reason or meaningful appeal is called Kafkaesque; the emoji-laden support replies are viewed as insultingly flippant.

Vendor lock‑in, “all‑in” cloud dependency, and victim‑blaming

• Some argue it was reckless to keep a “single copy” of critical data (photos, documents, credentials) in one proprietary cloud and treat an Apple ID as a “core digital identity.”
• Others push back: on mainstream platforms that dominate devices and services, this is “the main street, not a dark alley”; expecting non‑technical users to self‑host and design backup schemes is unrealistic.
• There’s recognition that “convenience as a drug” led many to accept walled gardens; several say this should be a wake‑up call that it “can happen to you.”

Gift cards, fraud, and AML

• Many suspect aggressive fraud or anti–money‑laundering (AML) systems were triggered by the high‑value gift card, noting gift cards are widely used in scams and laundering.
• Several describe known scams where physical cards are tampered with in stores, or where victims are forced to buy gift cards for scammers.
• Critics question why the entire Apple ID and devices are disabled instead of just blocking gift‑card use, calling it a “hammer to crack an egg.”
• Some resolve never to buy or redeem Apple gift cards; others note cards are often discounted or used to avoid storing card details with big tech, so the risk is non‑obvious.

Law, regulation, and recourse

• Strong calls for regulation: rights to data export on closure, transparent reasons for bans, and independent appeal/ombudsman processes, especially given IDs gate devices and sometimes government services.
• EU GDPR export rights and local civil/administrative tribunals (e.g., in Australia) are suggested as partial levers; others recommend demand letters or small‑claims actions to reach corporate legal teams.
• AML secrecy rules are cited as a possible reason Apple won’t explain the trigger, but several argue this doesn’t justify permanent, opaque lockouts of long‑standing accounts.

Backups, self‑hosting, and realistic mitigations

• Large thread on mitigation strategies: Time Machine with “download originals,” rsync/Arq to NAS or S3/Backblaze, Synology/Immich/Nextcloud/PhotoPrism, multi‑cloud mirroring (iCloud + Google Photos + OneDrive).
• Several note hard limits: iCloud “optimize storage” makes full local copies hard once libraries exceed local disk; backing up iMessages, shared iWork docs, and passkeys is especially tricky.
• Some argue 3–2–1 backup and avoiding single‑provider dependence is now essential; others say this is far beyond what average users can or will do, reinforcing the case for legal protections.

Platform power and broader implications

• Many generalize beyond Apple: similar horror stories from Google, PayPal, Amazon, banks; “live by Big Tech, die by Big Tech.”
• Concerns that government digital IDs and critical services increasingly depend on iOS/Android, amplifying the danger of unilateral “de‑platforming.”
• A minority advocate abandoning Apple/Google entirely in favor of Linux/BSD or smaller providers; others argue that, for most people and businesses, that’s not currently realistic.

Why email encryption lags behind HTTPS

• Commenters note most web traffic is HTTPS while email—often more sensitive—remains mostly unencrypted end-to-end.
• HTTPS became ubiquitous partly due to browser and search engine pressure; email has no comparable central push.
• Transport-level TLS between mail servers is now widespread, but end-to-end encryption is seen as breaking spam filtering, server-side rules, search, and especially webmail.
• Key discovery and cross-client support (S/MIME, PGP) remain fragmented and poorly standardized.

Usability, key management, and data longevity

• Personal key management across devices is widely viewed as the core unsolved problem: devices die, get stolen, or fall in lakes; users lose chat/email histories and keys.
• Some want a “super dumb, robust” multi-device key store; others suggest passkeys, hardware tokens (YubiKey-like rings/bracelets), or local password managers.
• There is tension between people who prioritize reliability and history vs. those who prioritize maximum security even at the cost of data loss.
• Losing access to S/MIME-encrypted email archives is cited as a real-world failure mode; some wish clients would store messages decrypted locally once received.

Threat models and tradeoffs

• One camp adopts a strong adversary model (states, pervasive surveillance) and accepts losing history as a feature.
• Another camp assumes weaker threats (random hackers, scams) and is willing to soften security to preserve archives and usability.
• It’s argued that if only “people with something to hide” use strong tools like Signal or PGP, they become easier surveillance targets; mainstream adoption matters.

Is encryption needed for email?

• Several see email as a “digital postcard,” mostly spam and notifications, fine without heavy crypto; for private messaging they prefer other tools.
• Others stress that people expect email to be private (password-protected accounts, sensitive content, receipts, logins), so default encryption would be safer than relying on users to choose.

Tools, providers, and ecosystems

• Mentioned tools: DeltaChat (moving away from classic email), mutt + GPG, Thunderbird, Mailvelope, Signal, WhatsApp, self-hosted bridges, password managers.
• Proton Mail draws criticism for limited interoperability and legal exposure, but others point out its public-key lookup endpoints do exist.
• An example from a small company and from government smart-card deployments shows S/MIME-by-default can work in controlled environments, albeit with search, webmail, and interop drawbacks.

What “skills” are

• Described as small, self-contained bundles: a SKILL.md with frontmatter (name + description) plus optional reference docs and scripts.
• On session start, coding agents scan skills folders and inject only the short descriptions into the system prompt; full content is lazily read when relevant.
• Many commenters frame this as “dynamic prompt/context extension” or “context-management for tasks,” often analogous to sub-agents or English header files.

Usage patterns and benefits

• Common uses: project-specific coding help, debugging flows, CI/build result retrieval, document editing, front-end design, generating charts via Python, and browser/RE tooling (Playwright, Ghidra).
• People like being able to codify repeatable workflows (“next time, just do this”) and keep them out of the main context until needed.
• A recurring pattern is having the LLM write or update skills itself, then lightly editing them. Teams see potential for shared skill libraries encoding house style, APIs, and processes.

Implementation & ecosystem

• Skills are supported in Codex CLI and ChatGPT’s code environment; Claude Code pioneered the pattern; Gemini and other tools are adding equivalents.
• Local LLMs can also drive skills if they have shell/file access and enough context for long tool-calling loops.
• Comparisons to MCP: MCP exposes big tool catalogs up front; skills are lighter, pay-per-use, and often built atop CLIs. Some see skills as a better default for many use cases, with MCP reserved for richer RPC-style integrations.

Simplicity, prior art, and complexity fatigue

• Several argue skills are just formalized prompt stuffing or “documentation for the AI,” not a fundamentally new invention; others counter that the specific packaging + lazy loading is a meaningful UX/architecture win.
• Some feel overwhelmed by yet another layer (agents.md, MCP, skills, AGENTS.md), while others praise skills as the simplest way so far to extend coding agents.

AGI and intelligence debate

• Long subthread debates whether developments like skills show we’re far from AGI (we’re hand-writing “library functions” in English), or whether LLMs already qualify as a form of AGI by technical definitions of “general intelligence.”
• Discussion covers benchmark overfitting, Goodhart’s law, human vs machine “understanding,” and whether “real intelligence” is even definable in a non-circular way.

Vendor strategies and concerns

• Anthropic is praised for “obvious in hindsight” abstractions (MCP, skills, Claude Code) and coherent framing; OpenAI is seen as quietly following with massive distribution.
• Some want standardized, cross-vendor skills (tied to AGENTS.md / Agentic AI Foundation); others note security pitfalls (especially with MCP) and the risk of misuse.
• A separate warning highlights ChatGPT’s effective input cap being lower than the advertised context window, causing silent prompt truncation.

macOS HDR Behavior

• Several commenters complain HDR on macOS looks “washed out” on third‑party HDR monitors (especially OLED): blacks become gray, SDR UI elements look flat, while HDR video in a window looks fine.
• Others say this is by design: macOS keeps UI in SDR while only HDR content uses extended range, and on limited‑brightness displays the SDR UI will appear gray compared to HDR highlights.
• There’s disagreement whether raised blacks are an intentional trade‑off, an Apple bug, or a calibration/metadata issue; Windows’ HDR calibration tool is cited as working better on the same displays.

What RDMA over Thunderbolt Enables

• Previously, people chained Macs using pipeline parallelism (layers split across machines). This allows larger models than fit on one Mac but doesn’t speed up inference.
• RDMA over Thunderbolt plus MLX now enables fast tensor/head parallelism: each layer is sharded across machines, with per‑node Q/K/V, local attention, then all‑reduce on outputs.
• Reported benchmarks: ~3.5× speedup in token generation on 4 machines for batch size 1, mainly from reduced per‑node memory bandwidth pressure. Latency and frequent synchronization remain the main challenges.

Mac Clusters vs GPU Rigs (Cost, Power, Memory)

• Enthusiasts see M‑series clusters as attractive “AI appliances” for labs, small shops, and serious hobbyists: huge unified memory, low power, plug‑and‑play, no CUDA.
• Critics argue Nvidia/AMD GPUs are far cheaper per FLOP and have much higher raw bandwidth; memory bandwidth and interconnect, not just capacity, are the real bottlenecks.
• One comparison:
  • $50k Mac Studio M3 Ultra cluster: ~3 TB unified memory, slow (15 tok/s) but can host ~trillion‑parameter models.
  • ~$50k RTX 6000 workstation: much higher tokens/sec but limited to <400B‑parameter models (384 GB VRAM).
  • Similar‑capacity GH200 setups cost an order of magnitude more.
• Others point out you can build used Xeon/GPU franken‑clusters or multi‑3090 rigs, trading efficiency for raw capacity and heat.

Thunderbolt/RDMA Technical & Physical Limits

• RDMA runs over Thunderbolt/USB4 PCIe (effectively PCIe 4×4, ~64 Gbps per port), lower latency than standard TB networking.
• Topology is a fully connected mesh; practical limit is ~6 Mac Studios, so this is not a large‑scale datacenter fabric.
• People worry about Thunderbolt’s mechanical robustness for semi‑permanent interconnects, but note locking USB‑C variants and third‑party “cable locking” accessories.
• Some lament the lack of Thunderbolt switches or QSFP‑style ports; others note a “Thunderbolt router” could just be a multi‑port computer.

Deployability and Server‑Style Management

• macOS is seen as awkward in a datacenter role: GUI‑driven OS upgrades, weaker open tooling vs Linux/BSD, no real IPMI/iLO equivalent.
• MDM‑based workflows (Jamf, open‑source MDMs, erase‑install scripts, VNC/Screen Sharing) can automate upgrades and remote control, but require Apple‑specific expertise.
• Rackmount concerns: Mac Studio’s rear corner power button and non‑locking TB cables make clean rack deployments fiddly; third‑party rack kits and locking accessories partly address this.
• Some miss Xserve and argue Apple has never fully committed to server‑grade macOS; others note AWS and MacStadium already run Mac fleets successfully.

Apple’s Possible Strategy and Ecosystem Play

• Several commenters see this as part of a broader Apple plan:
  • Bake AI accelerators and large unified memory into all high‑end Macs.
  • Make Macs attractive for AI research and local inference.
  • Potentially reuse the tech to distribute AI workloads across a user’s devices (Mac, iPhone, iPad, Apple TV, HomePod) for private, on‑prem inference.
• Others are skeptical: RDMA over Thunderbolt is limited to small clusters and doesn’t directly translate to Apple’s mostly wireless consumer device network.

Wider Market, RAM, and End‑User Impact

• There’s debate over whether high‑RAM Macs could become a cost‑effective medium‑scale inference platform, especially given current DRAM shortages and price spikes.
• Some fear that high‑end Macs will be bought out by commercial AI users; others reply that typical home users neither need nor can justify 512 GB+ Macs anyway.
• A long subthread argues over whether RAM pricing spikes are a short‑term bubble or a multi‑year structural issue, with implications for “a computer in every home” and for cheap local AI.

Security, Scope, and Miscellaneous

• RDMA is disabled by default and must be explicitly enabled in recovery mode, which alleviates some concerns about plug‑and‑play physical attack vectors.
• Not tied to ML: in principle any distributed workload that benefits from low‑latency, high‑bandwidth memory access could use it (e.g., MPI/HPC), though early tests are rough.
• Gaming and eGPU hopefuls ask if this helps them; consensus is no—this is for clustering Macs, not reviving general eGPU support or multi‑node gaming.

Ownership, Licenses, and “My Device, My Content”

• One side argues that once a reader has paid for a book, how they process it (including with an AI tool) is their business; authors “got their money” and shouldn’t control reading methods.
• Others push back that with Kindle it’s only a revocable license, not true ownership, and Amazon’s DRM means “my device, my content” is factually wrong; Amazon ultimately decides what features exist.

Fair Use, Legal, and Contract Questions

• Some commenters call the feature “perfectly reasonable fair use,” likening it to a bookstore clerk answering questions or a reader writing notes/reviews.
• Others emphasize scale and automation: an LLM operating over the entire Kindle corpus is different from individual human reading, and training vs. inference is a legal gray area.
• There’s concern about whether uploading text to servers counts as distribution and whether publisher–Amazon contracts allow this kind of processing.
• A few point out recent rulings suggesting that training on legally acquired works can be fair use, though details remain contested.

Technical Implementation and Training Concerns

• Many assume the system won’t locally run; questions arise whether Amazon is reusing publisher files or user uploads.
• Several note that LLMs can answer questions by putting the book (or the portion read so far) into the context window at inference time, which is distinct from training.
• Skeptics doubt Amazon won’t also use this data for training, given its track record, and suggest “poisoning” Kindle-only junk content to pollute models.

Reading Experience and Target Audience

• Enthusiasts see it as extremely useful: recaps after long breaks, tracking minor characters, understanding dense classics, long fantasy series, textbooks, and generating study questions.
• Others deride it as a crutch for people who “hate reading” or “can’t be bothered to read properly,” arguing that forgetting earlier details is part of normal reading, or that this outsources the core experience.
• Some emphasize that people with limited time, long/complex books, or kids and jobs may genuinely benefit, comparing it to fan wikis and glossaries.

Accuracy, Hallucinations, and Alternatives

• Skeptics cite Amazon’s own faulty AI recap of its Fallout TV show as evidence that such systems can misrepresent works, especially with minimal human oversight.
• Supporters counter that text-based book Q&A is easier than video recap and should be more reliable if grounded in the full text.
• Several say they’d still trust well-maintained fan wikis over LLM interpretations for plot details and canon accuracy.

Authors’ Role and Control

• Some argue authors have no say in how readers navigate their books, even if it “spoils” mysteries or structure; others note that many works are carefully crafted for linear discovery.
• There is criticism that authors/publishers weren’t notified and can’t opt out. Others frame that as acceptable: this is a reader-side tool layered on top of legitimately licensed content.
• One perspective from an author is that aggregated question data could be invaluable feedback on confusing or impactful parts of a book.

Overall reaction & novelty

• Many commenters find the project delightfully absurd, “cyberpunk,” and one of the best Show HNs lately.
• Others are conflicted or disturbed by the image of a rat in VR on a trackball with no easy exit.
• Several people note it resembles both sci‑fi jokes and real historical projects (e.g., pigeons guiding bombs).

Evidence of gameplay & missing video

• A recurring frustration is the lack of clear video of rats actually playing Doom on the current setup.
• Links are shared to older videos showing rats running down a straight corridor in Doom and a short clip of the newer rig, but no full gameplay session.
• The author explains the second‑generation rig took so long that the pet rats aged out; only habituation was done, not full Doom training.
• Hardware and software are open‑sourced, with encouragement for labs or hobbyists to continue the work.

Ethics & animal welfare

• Some are reassured that no surgery is involved and that this seems more benign than typical lab experiments.
• Others argue any non‑consensual animal experimentation is unethical, especially when reality is being altered via VR and the animal is physically constrained.
• Concerns about sugar‑water rewards are raised; doses are small, and alternatives (e.g., altered drinking water) are discussed.
• A few view it as no worse than pet training or work animals, and some even see it as enrichment if the rats enjoy the task.

Technical design, behavior, and suggestions

• Praise for the custom hardware, VR rig, and attention to whisker space; air puffs tied to in‑game collisions are noted as clever.
• Suggestions:
  • Release parametric CAD files and bill‑of‑materials cost estimates.
  • Adapt setups for mice, cats, or other species.
  • Reduce reward latency; use clicker‑style conditioning.
  • Better match rat vision: wider field of view, panoramic displays, possibly dual virtual cameras; some see current design as anthropocentric.
  • Alternative control schemes (chin/bite triggers) and first‑person games beyond Doom.

Doom as meme and platform

• Multiple comments explain Doom’s role as a historically important, mod‑friendly FPS and why “can it run Doom?” became a cultural meme.
• Jokes about future “running Doom on rats / rat brains” and rodent esports and warfare appear throughout.

Expanding Surveillance and Flock’s Role

• Many see the Florida law as effectively forcing citizens to submit to Flock’s ALPR/“vehicle fingerprint” system, which logs not just plates but make, color, stickers, damage, etc.
• Links are shared showing Flock data being used by local cops, federal immigration authorities, and even abused in domestic contexts (e.g., ex-partners).
• Some argue the data is broadly accessible via law-enforcement networks and partially by FOIA; others push back that “ANYONE” is an exaggeration but concede the access scope is still troubling.

Legality, Intent, and Rights While Driving

• Several contend that deliberately modifying a plate so cameras can’t read it is obviously illegal, akin to tampering with a passport.
• Others emphasize intent: random mud or defects vs a carefully crafted adversarial pattern specifically meant to defeat ALPR.
• Debate over “knowingly” in the Florida statute: does it require deliberate evasion, or merely awareness your plate is obscured?
• There is disagreement on “driving is a privilege”: some say that framing has eroded rights; others note courts have long tolerated reduced privacy for drivers while still requiring probable cause for searches.

Technical Efficacy and Countermeasures

• Skepticism that the adversarial pattern really works in the wild: angle, noise, model differences, and retraining could break or neutralize it.
• Discussion of alternative tactics: opaque or clear plate covers, mud, paint thinner, fake leaves, bike racks, IR lighting, and artistic wraps that confuse computer vision.
• Florida has newly outlawed most covers and frames, with fines and possible jail time; some note enforcement has historically been very lax.

Privacy vs Enforcement and “Nothing to Hide”

• One side worries mass plate tracking functionally recreates warrantless GPS tracking and dragnet searches that courts have otherwise limited.
• Others argue it’s just automating observation officers could make anyway, and that oversight and data controls matter more than banning the tech.
• A “nothing to hide” stance is voiced; critics respond that future political shifts could turn harmless data into a weapon against ordinary people.

Broader Authoritarianism and Exit Fantasies

• Multiple comments connect Flock, VPN bans, and similar laws to rising fascism/technocratic authoritarianism in the U.S.
• Some recommend guns, bug-out bags, and escape plans (walk to Canada/Mexico, passports, crypto); others call this survivalist fantasy and advocate focusing on elections instead.
• Florida specifically is described as increasingly hostile (politically, environmentally, educationally), prompting some residents to consider leaving.

Home Depot’s Response and Legal Caution

• Commenters are struck by Home Depot’s lack of communication with the researcher and press, interpreting the silence as legal/PR strategy once “the media” was involved.
• Some argue this is rational in a litigious, shareholder-driven environment, even if it prevents a transparent postmortem.

Customer Service and Store Experience

• Experiences with Home Depot staff vary widely: some report attentive help, others say employees are absent, disengaged, or lack basic tool knowledge.
• Comparisons: Lowe’s is often seen as marginally better; Ace/local hardware stores are repeatedly praised for knowledgeable “old hands” and human service.
• Several people now just order online for in-store pickup to avoid wandering large, understaffed stores.

Surveillance, Theft, and Local Economies

• Discussion branches into Flock license-plate cameras in big-box parking lots.
• One side emphasizes theft reduction; the other emphasizes privacy, anti-surveillance, and resentment of corporations “sucking towns dry.”
• Some distinguish anti-surveillance from “pro-theft,” and complain about bad in-store UX (locking items, friction-heavy rebates).

Website, Apps, and Internal IT Quality

• Many describe Home Depot’s website/app as slow, buggy, and poorly designed (random store selection, unusable mobile performance, broken filters/sorting).
• In-store connectivity is poor (steel “Faraday cage”), pushing people onto unreliable WiFi; carrier-managed auto-join networks and VPN incompatibility add friction.
• A minority defend the site’s inventory accuracy when it does load.
• Anecdotes about Home Depot’s modernization push (K8s/React, conference recruiting) suggest internal confusion, legacy systems, and lack of coherent strategy, contrasted with praise for Walmart’s modernization.

Token Exposure, Secret Scanning, and Risk

• Multiple commenters note GitHub’s and some AI providers’ secret-scanning that auto-revokes exposed keys, but say coverage is imperfect and usually limited to GitHub itself or main branches.
• It’s unclear from the thread where the Home Depot token was exposed; several assume it wasn’t in a public GitHub repo or it would’ve been caught.
• Potential damage discussed: cloning source code to mine for vulnerabilities and, if CI/deploy access existed, inserting malicious changes.

Broader Security Culture and Mitigations

• Debate over whether security “really matters” given mild market consequences for big breaches; others counter that huge effort prevents far more incidents.
• “Vibe coding” and poor key hygiene are seen as growing risks.
• Suggestions for self-hosted secret management include platform-native secrets, password managers with APIs, and tools like SOPS + age.

Unionization in Tech & Game Development

• Many see game devs as especially in need of unions: chronic crunch, unpaid overtime, mass layoffs after launches, and exploitation of “passion for games.”
• Others argue software engineers are already well-paid and comfortable, so unions may be a “luxury option,” though even skeptics acknowledge conditions in game studios can be harsh.
• Some point to industry models like Hollywood (project-based work with strong unions) as a possible future for tech and games.

Power, Leverage & Replaceability

• Debate over how much leverage software workers have: unlike factory labor, software keeps running for a while without its creators, which weakens strike power.
• Counterpoints: deep domain knowledge, old stacks, on-call duties, and brittle infrastructure mean a few key engineers are hard to replace; one bad deployment can have huge impact.
• Outsourcing and IP licensing are raised as theoretical union-busting tools, but commenters note that in practice replacing entire experienced game teams is risky and difficult.

Politics & Scope of Union Agendas

• One camp wants unions “monomaniacally” focused on wages, hours, and workplace issues, warning that taking positions on Gaza, BLM, etc. is divisive and weakens organizing.
• Another argues you can’t separate worker issues from discrimination and broader politics: unions represent all workers (including marginalized groups) and must defend them.
• Historical perspective: unions have often been key political actors against oligarchy; some think avoiding politics is naive given that employers are highly political.

Economics, Company Failure & Offshoring

• Concern that aggressive bargaining in a downturn can bankrupt firms, hurting workers; critics cite Yellow Trucking and offshoring of film work to Europe/Asia.
• Others counter that mismanagement and debt usually kill companies, not unions, and that businesses whose viability depends on exploitation “shouldn’t survive.”
• Broad discussion of market power: employers colluding on wages, wage theft, and concentration vs. unions as a partial counterweight.

Legal & Organizational Context

• Id’s union is with Communications Workers of America, under the AFL-CIO; described as a “wall-to-wall” industrial unit (everyone non-management in the studio).
• Comparisons to craft vs industrial unions and Hollywood contracts, where overtime multiplies rapidly and makes endless crunch expensive.
• Some note union strength in the U.S. depends heavily on the NLRB and administration; enforcement can be undermined politically.

Immigration & Labor Supply

• CWA’s attempt to challenge the OPT program is cited as an example of unions opposing mechanisms that can weaken bargaining power via cheaper, less-protected labor.
• Discussion of how current U.S. immigration regimes (e.g., H-1B, undocumented work) are used to undercut wages and keep workers too vulnerable to organize.

Alternatives & Tools

• A number of engineers say they’d prefer strong labor-law enforcement (hours limits, notice for layoffs, real overtime rules) over unions, but others note that this is exactly what unions historically fought for.
• Suggestions for “labor tech”: apps for documenting violations, connecting gig workers, and organizing securely outside employer-controlled channels.

Variable font features and flexibility

• Many commenters welcome another high-quality variable font, especially one that’s open and broadly usable on the web.
• Google Sans Flex’s multiple axes (weight, width, roundness, etc.) are praised as powerful; the roundness axis is seen as particularly uncommon and interesting.
• Some note that Google Fonts’ variable controls have improved and now allow fine-grained tweaking (e.g., font-stretch), but warn this can lead to over-optimization and time-wasting.
• Roboto Flex is cited as a benchmark, with more axes and even finer control, making Flex notable more for openness than for sheer capability.

Licensing and openness

• Compared with Apple’s San Francisco (seen as tightly and restrictively licensed), Google’s move is viewed as a positive step for designers and developers.
• People who have fought “font licensing hell” are especially appreciative of more redistributable, web‑safe options.
• However, the Open Font License–style “Reserved Font Name” clause means you can’t ship a modified version under the same name, so practical community collaboration on this exact font is limited.

Legibility and glyph disambiguation

• A large subthread criticizes Google Sans Flex for poor distinction between iIlL1 and 0O, failing the common “iIlL0Oo / i1IlL0Oo” test.
• Many argue that ambiguity is unacceptable for tokens, passwords, codes, and technical UIs; they advocate fonts with clearly distinct glyphs or slashed/dotted zeros by default.
• Alternatives praised for clarity include Ubuntu, Nunito Sans, IBM Plex Sans, Atkinson Hyperlegible (and its Next/Mono variants), and various monospaced programming fonts.
• Others counter that this is an interface/display font where context often suffices; for sensitive strings one can switch to a specialized font or avoid ambiguous characters entirely.

Aesthetics, use cases, and “does this matter?”

• Some find Google Sans Flex visually bland or “geometric” and less suitable for long body text, preferring Roboto, Inter, or classic humanist/grotesque sans‑serifs.
• Others argue fonts do matter for legibility, accessibility, and international coverage, but question the need for yet another near‑indistinguishable sans‑serif.
• There’s mild confusion about Google open‑sourcing a font closely tied to its brand, potentially diluting the visual distinction of its own products.

Role of Nuclear vs Renewables in Decarbonisation

• One camp sees nuclear as essential baseload to replace coal and gas, citing France’s much lower CO₂ intensity vs Germany, Poland, Italy despite all having substantial renewables.
• Others argue nuclear is “not key” but at best complementary: the real driver is rapidly falling‑cost wind, solar and batteries, already being deployed at far larger scale globally.
• China is used by both sides: some highlight its fast, cheap reactor builds; others note nuclear is a small, slowly growing share compared with explosive wind/solar growth.

Cost, Timelines, and Industrial Capacity

• Critics stress Western new‑build nuclear: very high capital cost, 15–20‑year lead times, and chronic overruns (European EPR projects, Vogtle, Hinkley Point C, planned EPR2).
• Pro‑nuclear replies blame FOAK designs, degraded supply chains, and hostile/chaotic regulation rather than intrinsic tech; point to much faster, cheaper builds in China and Japan.
• Debate over whether Europe has effectively lost its nuclear industrial base vs still having strong firms and expertise that could scale up again.
• SMRs are discussed as a way to standardise and factory‑build, but their eventual costs are viewed as highly uncertain and possibly over‑hyped.

Grid Integration, Intermittency and Storage

• Nuclear supporters emphasise dispatchability and high capacity factors; argue that intermittent renewables plus gas backup yield volatile prices, heavy fossil subsidies, and pollution.
• Renewable advocates counter that new wind/solar are far cheaper per GW and far faster to deploy; storage and grid expansion (including hydrogen‑ready gas plants) are seen as the real bottlenecks.
• There is disagreement on how far renewables plus storage can scale before hitting hard limits in northern Europe’s climate.

Safety, Waste, and Environmental Impacts

• Several argue that, even including Chernobyl and Fukushima, nuclear is among the safest energy sources; fears are labelled “radiophobia”.
• Others focus on low‑probability catastrophic risks and millennia‑scale waste, and point to environmental damage from uranium mining.
• Counter‑arguments note the small physical volume of spent fuel and the substantial, often ignored waste and pollution from fossil fuels and also from renewables (e.g., turbine blades, panel disposal).

Politics, Security, and Manipulation

• Commenters tie European gas dependence (especially on Russia) to nuclear phaseouts and see nuclear fuel as more secure due to diversified uranium supply and stockpiling.
• Opponents highlight links between nuclear and military or national‑prestige agendas, plus corruption and poor governance (e.g., Fukushima decisions).
• Some suspect heavy online lobbying and information operations on all sides—fossil, nuclear, and renewables—making honest debate difficult.

Enterprise dissatisfaction and migrations

• Many commenters report strong negative sentiment toward Oracle in their organizations, often citing “predatory” behavior, aggressive audits, and licensing/Java subscription tactics.
• Several enterprises are actively trying to reduce or eliminate Oracle, typically by:
  • Building institutional experience with PostgreSQL or other DBs at the departmental level first.
  • Migrating low‑risk or “low‑hanging fruit” workloads, then gradually larger systems.
  • Using AWS RDS or similar to add an abstraction layer and weaken the direct Oracle relationship.
• Some vendors are dropping Oracle support and providing migration paths (often to SQL Server), reinforcing the “trend away.”
• One anecdote describes a large bank slowly decoupling from Oracle middleware/DBs as part of a broader modernization and Sun hardware retirement.

Difficulty, risk, and economics of switching

• Multiple replies stress that moving off Oracle is extremely hard, expensive, and politically risky, often taking many years and risking failed or partial migrations.
• Even if Oracle is expensive, the migration cost and risk (including potential outages and dual‑system operation) can dwarf annual savings.
• Others argue that in some cases the cost savings are so large (e.g., many millions per year) that the risk is justified, though these are framed as exceptions requiring long, staged efforts.

Contract structure and renewal dynamics

• Discussion around 10‑year “unlimited” enterprise agreements:
  • Not typical for most software, but plausible for very large infrastructure deals.
  • These can feel cheap upfront but lead to painful “true‑ups” later as usage grows.
  • Some enterprises have used the 10‑year window to move workloads off Oracle before renewal to gain leverage.
• Skeptics question how widespread such contracts are and request sources; others note that Oracle has weathered similar “we’re moving off” talk for decades.

Oracle vs. competitors (SAP, PostgreSQL, SQL Server)

• One story: after an RFP process, a company chose SAP over Oracle due to concerns about Oracle’s honesty and behavior, despite SAP’s own issues; HANA itself then created technical and cost problems.
• Some technical managers still reflexively view Oracle as the only “real” enterprise DB and distrust PostgreSQL, especially where commercial Postgres support is weak.
• Others argue almost no one picking a greenfield system today chooses Oracle DB unless pulled in via an Oracle/SaaS application.

Java, Oracle, and developer sentiment

• Strong disagreement on whether Oracle “destroyed” or “rescued” Java:
  • Some say post‑acquisition years were worrying but Java is now in its best technical shape (JDK pace, performance, tooling).
  • Others claim Oracle’s licensing and lawsuits eroded goodwill, and that few new developers choose Java voluntarily, even though it remains ubiquitous in big enterprises.
• Several note that large-scale high-performance systems (e.g., Netflix-like workloads) still lean heavily on Java/HotSpot; newer languages may have more “hype” than actual deployment share.

AI/OpenAI bet and market risk

• Some commenters say headlines misstate the situation, arguing Oracle is being paid huge sums by OpenAI rather than “betting” $300B on it.
• Others point to credit default swap pricing and debt market signals indicating Oracle’s AI-related capex and contracts are seen as risky, though not catastrophically so.
• One view: even if many AI projects fail, Oracle’s “support and lock‑in” model might profit from enterprises deploying fragile AI systems that need intensive care.

Why Oracle persists

• Reasons given for continued Oracle use:
  • Legacy decisions with deep lock‑in (proprietary extensions, decades of apps built around Oracle).
  • Enterprise features, certifications, compliance, and “one throat to choke” support.
  • Government and large corporate procurement habits; Oracle often bundled with major ERP/financial packages.
• Some users praise Oracle Cloud’s pricing and footprint (while still avoiding Oracle DB itself on that cloud).

Article & Title Reception

• Many feel the post underdelivers on its headline; it’s mostly a report on workshops, with little actual math or technical depth.
• The original “AI systems don’t think” framing is seen as provocative and distracting; some welcome the later, softer rewording, others say the article still leans too hard on that claim without defining “think”.

Teaching AI with School Maths

• Several commenters like the idea of using ANNs/ML examples to teach secondary-school math and demystify AI.
• Others criticize the chosen examples (e.g. traffic-light classification) as unrealistic or conceptually sloppy, and hope future curricula will use better-grounded problems.
• Some wish they’d been taught neural nets earlier, contrasting this with older AI courses that dismissed ANNs in favor of other methods.

What Does It Mean for AI to “Think”?

• A long thread debates definitions: thinking vs reasoning vs computation vs consciousness.
• One view: AI is “just maths” and computation; humans are “just biology/physics”; in both cases that doesn’t settle whether there is “thinking” or consciousness.
• Another view: without a clear, testable definition of “thinking”, blanket claims (“AI does/doesn’t think”) are unfalsifiable and mostly rhetorical.
• Functionalist, substrate-independent positions (brains and computers can realize the same processes) clash with views that brains do something qualitatively different or not yet mathematically formalized.

Turing Test, Chinese Room, and Thought Experiments

• Some argue that modern LLMs can effectively pass Turing-like tests, at least over finite conversations; others say it’s still easy to expose them if you know what to probe.
• The Turing Test is criticized as less discussed just as systems become competitive at it.
• The Chinese Room thought experiment is revisited: some see it as useless or question-begging; others see it as a live challenge to claims that symbol manipulation equals understanding.
• Pen-and-paper and brain-simulation arguments (Church–Turing, simulations vs reality, map vs territory) lead to disputes about whether simulating a brain would yield genuine consciousness.

Limits, Capabilities, and Anthropomorphism

• Capabilities cited: strong performance in math/programming contests, code generation, few-shot learning in-context, emergent computation inside transformers.
• Limits cited: inability to robustly correct its own reasoning, dependence on training distributions, hallucinations, high energy use, weak arithmetic without tools.
• Some argue anthropomorphic language (“LLMs are dishonest”, “they believe…”) and commercial AI hype mislead the public into over-ascribing agency or thought.
• Others argue that, regardless of labels, these systems already match or exceed humans on many tasks, and the human–machine gap may be narrower than people want to admit.

Article accessibility and exposition

• Several readers found the Quanta piece off-putting for “speedrunning” graduate-level prerequisites (manifolds, Hodge diamonds) before getting to the new result, even when they had the background.
• Others appreciated that someone tried to write about such a deep, technical result at all and thought the intro material on manifolds/rational parameterization was quite nice, even if the later parts became incomprehensible.
• Some commenters wished for more “explain like I’m 5” treatment of key concepts (e.g., Hodge diamond, mirror symmetry), similar to Wikipedia-level exposition.

String theory: framework, predictions, and value

• One camp sees string theory as good at generating rich mathematics (mirror symmetry, AdS/CFT, holography, etc.) but poor at producing concrete, testable physical predictions, especially compared to simpler theories.
• Others argue that any “theory of everything” will inherently be hard to test because relevant energies are far beyond current experiments; this is a domain problem, not specifically a string theory flaw.
• Several comments stress that string theory is better viewed as a framework compatible with many possible universes, not a single predictive physical theory; this broad compatibility is itself a reason it currently makes no sharp predictions.
• There is debate over whether a ToE should at least “retrodict” known results (e.g., hydrogen spectrum) and whether string theory has reached that bar.

Quantum gravity and testability

• Suggestions for where a unified theory might be testable: black hole horizons and Hawking radiation, extreme astrophysical environments, or subtle effects where GR and QM intersect.
• Others note many of these phenomena can be treated without full quantum gravity, and that quantizing gravity works in many regimes but breaks down at very high energies (near singularities or the Big Bang).

Cost and opportunity cost of string theory

• A back-of-the-envelope estimate puts four decades of string theory work at roughly $500M in salaries; some question if that’s worth it given limited physical payoff.
• Many argue this is modest compared to large experimental projects or even blockbuster movies, and that high-risk theoretical research is exactly what research funding is for.
• Counterpoint: the real cost is human capital—hundreds or thousands of very talented people may be “nerd-sniped” by a potentially unsolvable or non-physical program.
• Response: training people on hard, frontier problems has systemic value; most PhDs leave for other fields (e.g., finance, industry) where their skills still benefit society.

Mathematical content and how much credit string theory deserves

• The paper itself was linked, and commenters pointed to the Hodge diamond on page 6 as a central geometric object.
• Some push back on framing this as “string-theory-inspired”: Hodge structures and Hodge diamonds are standard in geometry and predate string theory; mirror symmetry and Gromov–Witten theory have string-theory roots, but much of the machinery is now mainstream math.
• Quanta’s implication that the proof “relies on ideas imported from string theory” was seen by some as overselling the string-theory connection when “differential geometry” might be more accurate.

Formal verification and computer-assisted mathematics

• Multiple commenters argue that, by 2025, major results should ideally ship with machine-checkable proofs (Lean, Coq, Metamath, etc.), which would save experts huge time and reduce reliance on informal reading groups to validate correctness.
• Others respond that, at the current frontier, fully formalizing a deep proof is extremely difficult and often takes many times more work than writing the informal proof; existing proof assistants and libraries are still immature for this scale.
• There’s a distinction drawn between:
  • Verifying a formal proof (easy for a machine once written), and
  • Translating an informal, intuition-heavy proof into a fully formal one (often years of work, even when the proof is well-understood).
• Some suggest future workflows where AI helps identify fragile steps or sketches formalizations without doing full end-to-end checking.

Meta-discussion and analogies

• A side thread compares funding string theory to funding speculative projects like Mars colonization or asteroid mining. Opinions range from seeing such ventures as inspiring and worthwhile frontiers, to viewing them as wasteful prestige projects with little societal benefit.
• Another subthread critiques HN culture: the tendency toward glib “just do X” prescriptions (e.g., “just formalize the proof”) and status-seeking via confident overstatements, versus recognizing the genuine difficulty of frontier work.

What the ruling actually does

• Commenters note both Ars and Reuters describe the same 9th Circuit order.
• Key changes vs earlier orders:
  • Apple may charge a “reasonable” fee for external payment links, tied to actual coordination costs and some IP compensation, but not to security/privacy costs.
  • External-payment buttons must be allowed but can’t be made more prominent than Apple IAP; Apple can no longer force them to be less prominent.
  • “Scare screens” about external payments are banned, but some exit screens are allowed.
  • Apple can again exclude certain categories (e.g. news/video partners) from using external links.
• Some see Epic’s spin (“end of the Apple tax”) as misleading, calling this a mixed or even mostly Apple‑favorable outcome.

Debate over “reasonable” fees

• Many expect Apple to push for a revenue percentage (though lower than 27%); others think a pure revenue cut conflicts with the “cost-based” language.
• Some argue Sweeney’s idea of “tens or hundreds of dollars per update” is unrealistic and would still exclude small developers if per-update.
• Several emphasize the court’s language that security/privacy costs can’t be used to justify the fee.

Walled garden vs user freedom

• One side: Apple’s lock‑down is harmful, especially on expensive iPads that can’t run third-party browsers or full dev tools; app review quality has degraded while scams and gambling apps proliferate.
• The other side: users knowingly choose the walled garden; it benefits non‑technical users and families, and Android or other devices remain alternatives.
• Disagreement over whether “you chose the garden” is meaningful when iOS/Android form a de facto duopoly and phones are societally mandatory.

Consoles, Google, and other platforms

• Some question why consoles (Sony/Microsoft/Nintendo) aren’t held to similar standards; replies argue:
  • Consoles are bought primarily as gaming devices, not general‑purpose tools.
  • They don’t have the same societal centrality as phones.
• Others note Epic has sued Google and Apple elsewhere; Epic’s console relationships and negotiated deals may be more favorable.
• Multiple comments flag Google’s upcoming Android “user choice billing” policies as mimicking Apple’s earlier 27% approach, with US carve‑outs due to prior litigation.

Developers, small vs large

• Some argue this primarily benefits large firms (Epic, Netflix, Spotify); many small developers may stick with in‑app purchase due to:
  • Comparable or higher all‑in costs for Stripe/Paddle plus tax compliance.
  • Lower conversion rates on external payment flows.
• Others counter that any ability to bypass 15–30% is especially important for small SaaS and game studios; more margin → more competition.

Security, fraud, and review

• Disagreement on how much Apple actually does to prevent scams:
  • Anecdotes of fraudulent apps (including password managers, gambling, “AI” clones) passing review, and superficial testing.
  • Others argue verification and ongoing fraud monitoring are non‑trivial and justify some cost.
• Some suggest simpler models (chargebacks, banning offenders) over pre‑emptive heavy control.

App stores, ownership, and reprogramming

• Several contributors want the right to install any software on owned devices, likening Apple’s control to a grocery chain selling a fridge that only accepts its own products.
• Others reply that manufacturers can define product constraints; if buyers accept them, that’s market choice, not coercion.
• Wider philosophical thread: programmable devices (phones, consoles, cars, radios, medical devices) should be user‑reprogrammable vs. safety/regulatory arguments for locking down certain categories.

Drivers of the DDR5 price spike

• Commenters report DDR5 kits up 200–300% since mid‑2024; some DDR4 has also tripled, including used modules.
• AI companies, especially OpenAI, are blamed for locking up a large share of global DRAM/HBM wafer capacity via multi‑year contracts, pulling supply out of the consumer and general enterprise market.
• Crucial’s exit from consumer RAM is framed as a symptom of Micron and others prioritizing hyperscalers over retail/OEM channels.

DDR4, trade policy, and supply constraints

• One line of argument: normally older DRAM fab equipment is resold to “budget” manufacturers (often China‑adjacent) to keep legacy DDR4 cheap while big players move to DDR5.
• Due to fear of US sanctions/tariffs, Korean firms allegedly are warehousing old tools instead of selling them, shrinking DDR4 capacity even as DDR5 remains tight.
• Others push back that DDR4 isn’t a true substitute for DDR5 for new systems, but concede DDR4 prices have still spiked for users trying to extend older platforms.

AI hyperscalers, legality, and antitrust

• Many see OpenAI’s behavior as an attempt to corner supply and raise rivals’ costs, likening it to classic market‑cornering schemes and calling for antitrust action or even criminal penalties.
• Counter‑arguments: large buyers reserving future capacity (like Apple with TSMC) is common; intent and actual non‑use of the memory would matter for any legal case.
• There’s disagreement over whether this already “clearly” violates monopsony/price‑discrimination laws or is just aggressive but legal procurement.

Impact on Framework, OEMs, and buyers

• Framework is seen as a downstream victim passing through costs; it has raised DDR5 prices ~50% and tightened return rules to prevent arbitrage (buy laptop + cheap RAM, return laptop).
• Industry reports cited show Dell, Lenovo, and Apple raising DRAM/NAND pricing 50%+ into 2026, with notebook shipment forecasts revised downward.
• Some individuals and companies are delaying hardware refreshes; others feel they “escaped” by buying high‑capacity RAM earlier and consider reselling or hoarding.

How long will this last?

• Several posts cite manufacturer guidance suggesting elevated prices until around 2028 due to slow capacity ramp‑up and uncertainty about an “AI bubble.”
• Others expect the usual boom‑bust “pig cycle” to eventually bring prices down via overcapacity or a demand crash, but the timing is considered highly uncertain.

Liberty, Liberalism, and Historical Lessons

• Several comments frame this as part of a long arc: classical liberalism and broad freedoms are seen as a rare, fragile historical exception that can easily regress.
• Others counter that liberal democracy and human-rights-based orders are currently the global norm and demonstrably successful; treating them as “dreams” is seen as internalizing anti-freedom rhetoric.
• Some adopt a pessimistic stance: keep fighting for liberty, but organize your life assuming things will keep getting worse and state power will grow.

Scope and Mechanics of the New Powers

• Key elements highlighted: state-developed spyware (“trojans”) to intercept encrypted communication; secret entry into homes if remote deployment fails; bodycams activated in homes when officers perceive risk to life or limb.
• A central unresolved question: are these always court-ordered, or can they be used without warrants? The linked article doesn’t mention “warrant,” which some find worrying and others call a reporting gap.

Security, Terrorism, and Foreign Threats

• Supporters emphasize real terror attacks and ongoing plots in Germany, plus aggressive foreign intelligence and “hybrid warfare” operations, arguing Europe can’t afford 2000s-style idealism.
• Critics respond that the terror threat is serious but not “existential,” and doesn’t justify extraordinary erosion of rights.

Slippery Slope and Turnkey Totalitarianism

• Many fear a familiar pattern: measures start for “extremist terrorism,” then expand to serious crime, then petty offenses, then political dissent.
• Some explicitly invoke historical German surveillance states and “turnkey totalitarianism,” warning that the same tools will be used by future illiberal governments and against those first frightened into accepting them.
• A minority dismiss slippery-slope worries as overblown but are challenged with examples of intelligence overreach and mission creep.

Legal Culture: Germany vs US

• One thread notes Germany’s traditional “inviolability of the home” and strict privacy norms; this kind of secret home entry to plant bugs was historically taboo.
• Others note the US has long allowed covert entries and technical surveillance with judicial orders, but emphasize system differences (e.g., no German-style exclusionary rule).

German Political and Cultural Context

• Some blame Berlin’s electorate for voting in a more conservative government after years of left rule; others argue Berlin is “drowning in crime” and welcome tougher policing.
• A side discussion portrays German culture as highly rule-bound and deferential to procedure, which some see as fertile ground for authoritarianism, though Germans in the thread dispute how universal that is.

Broader Anxiety and Systemic Pressures

• Several comments tie expanded surveillance to elite fear: sovereign debt crises, stagnant growth, climate breakdown, wars in Europe, rising protests, and potential revolutionary anger.
• The idea is that as power structures feel more precarious, they seek stronger tools to preempt unrest.

Repression, Speech, and Banking

• Some claim a wider European drift toward illiberal practices: debanking of disfavored activists, harsh policing of protests, and speech restrictions around Russia and Israel.
• They argue that once labeled (e.g., “Putin sympathizer” or “antisemite”), individuals can be targeted with both state force and private-sector sanctions.

Indexing JSON in SQLite: generated columns vs expression indexes

• Several commenters note that SQLite already supports “index on expression”, e.g. creating an index directly on json_extract(...), which can avoid generated columns entirely.
• Others argue generated (virtual) columns are safer and more self-documenting: they guarantee the index is used, whereas expression indexes are fragile and can be bypassed by small query changes (different JSON operators, quoting, or path syntax).
• Some point out that views plus expression indexes can give the convenience of columns without materializing them, similar in spirit to the article’s technique.

Reliability, schema design, and constraints

• Using JSON with generated columns is seen as common practice in multiple databases (SQLite, Postgres, SingleStore, SQL Server pre-native-JSON).
• Generated columns (or expression indexes) make it possible to index JSON fields and even enforce foreign key constraints when keys are buried inside JSON, though this may still require separate columns in some systems.
• A few commenters like the pattern of exclusively querying via computed columns so it’s impossible to accidentally write an unindexed JSON query.

JSON vs normalized data

• There is tension between fully normalizing data and using JSON(B) columns.
• Critics of JSON columns highlight difficulty in indexing, enforcing constraints, handling schema migrations, and potential overhead compared to normalized tables.
• Defenders argue JSON shines when:
  • Data is tree-shaped or highly nested.
  • External APIs or heterogeneous record types would require many relational tables.
  • Application code has richer type systems and handles migrations (e.g., via Zod or language-level types).
• Some describe hybrid models: core relational columns plus a JSON “bag of attributes” for rarely queried or highly variable fields.

Performance, alternatives, and missing evidence

• Commenters appreciate the trick and its simplicity, especially for small or embedded apps graduating from raw JSON files to SQLite.
• One person questions the title’s “full speed” claim, noting the lack of explicit benchmarks or query plans in the article.
• DuckDB is raised as a better fit for heavy analytical workloads over large JSON datasets, while SQLite remains favored for embedded and systems use.
• There’s interest in more powerful JSON indexing (e.g., multi-valued indexes for JSON arrays) and Postgres-style GIN indexes; SQLite currently lacks these.

Related formats and systems

• Comparisons are drawn to older XML “document store” databases and to MongoDB / JSONB-style storage.
• A side thread discusses Lite³, a serialized B-tree format for JSON-like data that supports zero-copy queries and in-place updates, contrasted with Postgres JSONB’s immutable, server-bound design.

Core approach: Treat every TV as a dumb display

• Many commenters say the real answer is simple: never connect the TV to the Internet and use HDMI from another device (PC, streaming box, console).
• Several report that recent LG, Samsung, Philips, TCL, Roku TVs work fine offline, though some brands/models nag about Wi-Fi or TOS repeatedly.
• A few people physically remove Wi-Fi modules or would do the same for future cellular modems.

Escalating tracking and “spy rectangle” fears

• Strong concern that TVs will eventually auto-connect via neighbor Wi-Fi, public hotspots, or embedded 4G/5G (especially with cheap 5G RedCap-style IoT modems).
• Anecdotes: insurance “wellness” devices and smart toothbrushes shipped pre-paired, silently uploading data; this makes people wary of any networked appliance.
• Some suggest honeypot open Wi-Fi or Faraday-cage-level defenses; others warn against repurposing embedded SIMs due to extreme overage charges and legal risk.

“Dumb” TVs and criticism of the article

• The article’s listed dumb TVs are criticized as low-end: HDMI 2.0, 4K/60 only, weak panels, poor sound, short warranties.
• Several say: buy the best smart panel you can, then keep it offline and ignore the built-in OS.
• The piece is called poorly researched (brand ownership errors) and overly slanted toward an Apple TV solution.

Apple TV vs TV spyware vs other boxes

• Debate: some see Apple TV as a practical “least bad” box with no OS-level ads; others argue its data collection is still substantial and undercuts the privacy framing.
• Alternatives mentioned: Android TV/Chromecast boxes, Nvidia Shield, Raspberry Pi/Kodi, Jellyfin/“high seas,” used last-gen consoles, HTPCs with keyboard/trackpad.
• One camp loves HTPCs for full browser access and flexibility; another finds “use a computer” a clunky living-room UX.

Blocking and hacking approaches

• Pi-hole/AdGuard and DNS interception can reduce some tracking, but can’t stop GUI ads, “suggested” content, or devices with hardcoded DNS/DoH.
• Jailbreaking LG TVs (rootmy.tv and successors) is praised for ad-free apps, remote remapping, ambilight, etc., but most easy exploits are now patched and fragile.

Broader “smart everything” backlash

• Parallels drawn to cars with telematics: some prefer pre-2014 vehicles or pull telematics fuses.
• Overall sentiment: best long-term pattern is to own the display, own the smarts, and minimize what any single vendor can see.