0:00 / 0:25ai Mistral’s Leanstral 1.5 pushes open AI toward machine-checked proofs
Mistral released Leanstral 1.5, an Apache-2.0 model for Lean 4 proof engineering with 119B total parameters and 6B active parameters. The company reports major benchmark gains, including saturating miniF2F, solving 587 of 672 PutnamBench problems, and achieving stated state-of-the-art results on FATE-H and FATE-X. Mistral also says the model helped verify code properties and found 5 previously unknown bugs across 57 tested repositories, positioning formal verification as more practical for real software work.
Discussion: Mixed — HN is broadly interested in the release, especially the idea of a relatively small, open, specialized model that can help with Lean proof engineering. But the discussion is skeptical of some marketing claims: commenters challenged the bug-finding example, questioned comparisons against older models, and emphasized that formal verification still requires users to understand the properties being proved. (Enthusiasm for specialized, low-cost, open models, Skepticism about the bug-finding and fuzzing claims, Interest in practical formal verification workflows)
0:00 / 0:24ai Codex users spot a 516-token reasoning cliff
A public Codex GitHub issue reports an apparent GPT-5.5-specific anomaly: responses clustering at exactly 516 reasoning output tokens, with smaller spikes around 1034 and 1552. The author says they analyzed 390,195 token records across 865 sessions from February through June 2026, and found GPT-5.5 accounted for 82% of exact-516 events despite being 19.3% of responses. The issue does not claim proof of hidden chain-of-thought truncation, but asks OpenAI to investigate whether a reasoning budget, routing, fallback, scheduler, or related behavior is causing premature stops on complex tasks.
Discussion: Negative — The HN thread is mostly worried and frustrated: commenters say they have seen Codex quality drops, some claim they can reproduce the 516-token pattern, and many frame it as a trust problem with closed, server-side model changes. A minority pushes back that the evidence may be incomplete or that local/open models have their own configuration pitfalls. (frustration with perceived coding-model regressions, concern over adaptive reasoning budgets and silent routing changes, requests for reproducible telemetry and public acknowledgement)
0:00 / 0:23ai AMD inference gets cheaper per token, but HN wants the watt bill
Wafer says it served GLM-5.2 on AMD MI355X hardware with MXFP4 quantization and reached 213 tokens per second in a single-stream test, plus 2,626 aggregate tokens per second per node at 2.4 requests per second on a long-context workload. The company argues this is about 80% of its measured B200 performance while using much cheaper AMD GPUs, and says the remaining gap is increasingly about framework support rather than an unbridgeable CUDA moat. The work involved choosing sglang, fixing speculative decoding support on ROCm, and tuning MoE kernel selection for GLM’s FP4 shapes rather than writing custom kernels.
Discussion: Mixed — Commenters were interested in AMD as a real alternative to scarce and expensive Nvidia capacity, but the thread was notably skeptical of the framing. The biggest objections were that performance-per-dollar is incomplete without power, cooling, and datacenter constraints, and that FP4 quantization may reduce model quality in ways benchmark tables do not fully capture. (Demand for performance-per-watt, not just performance-per-dollar, Skepticism about headline benchmarking and cherry-picked cost framing, Concern that FP4/MXFP4 quantization can degrade frontier-model quality)
0:00 / 0:39ai Newer Claude models stumble on stricter tool schemas
Armin Ronacher reports that newer Anthropic models, specifically Opus 4.8 and Sonnet 5 in his tests, sometimes call Pi’s nested edit tool with correct replacement text but extra invented fields that violate the schema. He argues this may be a post-training artifact: if models are optimized inside a forgiving Claude Code-like environment, they may learn habits that fail in stricter third-party harnesses. Strict tool invocation eliminated the issue in his runs, but the broader concern is that tool schemas are becoming distribution-dependent rather than neutral contracts.
Discussion: Mixed — HN largely found the write-up technically useful and plausible, but the discussion split between concern over hidden, model-specific harness behavior and pragmatic fixes like better validation errors, retries, hooks, and self-healing tool adapters. Several commenters saw this as an emerging compatibility problem for agent runtimes, while others treated it as normal engineering tradeoff territory rather than a crisis. (Helpful error messages and retry loops can mitigate malformed tool calls, Strict or grammar-constrained decoding versus possible quality and cost tradeoffs, Concern that post-training on Claude Code-like harnesses may create lock-in or brittle transfer)
Quanta reports that JWST’s view of the first billion years is forcing astrophysicists to revisit models of early black holes, galaxies, and mysterious “little red dots” that appear hundreds of millions of years after the Big Bang. Proposed explanations include gas-shrouded black holes or “black hole stars,” super-Eddington feeding, direct-collapse black-hole seeds, and early galaxies forming stars more efficiently or in bursts. The big takeaway is not that cosmology is broken, but that Webb has exposed a much more diverse and complicated cosmic dawn than researchers expected.
Discussion: Positive — HN’s mood is largely excited and curious: commenters are treating the Webb findings as a reminder that better instruments make the universe stranger, not simpler. There is some healthy skepticism and correction around possible brown-dwarf contamination, plus lots of speculative physics tangents and pop-science book recommendations. (Awe at JWST reshaping early-universe astronomy, Interest in little red dots and black-hole-star ideas, Skepticism about how much is signal versus contamination or interpretation)
0:00 / 1:26software C&C Generals runs on iPhone and iPad, but HN says the headline overclaims
A GitHub project has Command & Conquer: Generals Zero Hour running natively on Apple Silicon Macs, iPhone, and iPad, with touch controls and no bundled game assets. The project is built on EA’s GPLv3 source release via GeneralsX, which the README says already did the major macOS/Linux porting work; this fork adds iOS/iPadOS support plus engine fixes. The rendering path still goes through DirectX 8, DXVK, Vulkan, MoltenVK, and Metal, and users need to provide their own copy of the game assets.
Discussion: Mixed — HN was impressed that a 2003 RTS can run on Apple Silicon Macs and iOS devices with touch controls, and many saw it as a strong example of human-directed AI-assisted porting. But the dominant pushback was that the title is misleading: commenters repeatedly noted the README credits EA’s GPLv3 source release and fbraz3/GeneralsX for the macOS/Linux heavy lifting, while this fork adds iOS/iPadOS support and fixes. (Skepticism about the title and scope of Fable’s contribution, Interest in AI-assisted porting and game preservation, Debate over reverse engineering, source releases, and piracy boundaries)
0:00 / 0:20software A tiny rotate button exposes a big UX rule
The article compares photo rotation on an iPhone and a Nothing Phone: rapid repeated taps on the iPhone are remembered, while on the Nothing Phone/Android the button gives haptic and sound feedback but ignores taps during the rotation animation. The author argues the core failure is making users wait for decorative or transitional animation, especially in workflows where a supposedly casual control becomes repetitive work. The broader point is that interfaces should either buffer, interrupt, or accelerate state changes—but not confirm input and silently discard it.
Discussion: Mixed — Commenters broadly agreed that giving haptic or visual feedback and then ignoring the tap is a trust-breaking interaction. But the thread pushed back on the absolutism of “a button has one job,” with debates about debouncing, accidental double taps, accessibility needs, and when queued actions can be worse than dropped ones. (feedback must match accepted input, animations should not block user intent, debouncing versus buffering)
0:00 / 0:20software A Linux htop explainer gets a second life↻ from 2019
This 2019 explainer resurfaced on HN today, walking through what htop and top display on Linux and where those numbers come from. It uses procfs and strace examples to unpack uptime, load averages, task and thread counts, PIDs, process trees, users, and why load average is not simply CPU usage. The value is less breaking news than durable operational literacy: understanding these fields helps diagnose real Linux performance issues without guessing.
Discussion: Positive — The thread is warmly appreciative of the explainer, with several readers saying it clarified tools they use regularly. Discussion broadened into practical habits and alternatives like btop, nmon, and procs, plus detailed debates about memory metrics, GPU visibility, and terminal UI tradeoffs. (Evergreen Linux fundamentals still valuable, Practical htop configuration tips, btop and other monitoring-tool comparisons)
0:00 / 0:28software ProseMirror’s creator plants a new rich-text editor
Wordgard is a new MIT-licensed JavaScript library for building in-browser rich-text editors, positioned as a structured, schema-based system rather than a free-form HTML editor. It advertises modular extensions, accessibility, right-to-left support, structured content like tables and nested lists, and collaborative editing. The HN discussion treats it as a serious successor or sibling to ProseMirror, but developers are asking what justifies the migration cost and how it will stack up against options like Lexical.
Discussion: Mixed — The thread is broadly impressed and respectful, especially toward Marijn Haverbeke’s track record and the site’s artwork, but the developer reaction is not uncritical. The main hesitation is practical: why switch from ProseMirror, TipTap, or Lexical, especially with no easy migration path and early mobile issues reported by commenters. (Admiration for ProseMirror and confidence in the author’s craft, Questions about the rationale and switching cost versus ProseMirror or Lexical, Interest in schema design, extension APIs, collaboration, and typed document representations)
0:00 / 0:21software Shadcn/UI makes Base UI the default, but says Radix isn’t going away
Shadcn/UI now defaults new projects to Base UI instead of Radix, citing Base UI’s stability, ongoing component work, internal use, and a claimed 2-to-1 preference in new shadcn/create projects. Radix is not being deprecated: the project says updates and new components will continue to ship for both libraries where possible, and existing Radix users do not need to migrate. The changelog also introduces an agent-oriented migration “skill” for projects that do want to move component by component.
Discussion: Mixed — HN’s reaction was split: the announcement drew interest because shadcn/ui is widely used, but much of the top discussion fixated on the post’s perceived AI-written marketing tone rather than the component-library change itself. On the substance, commenters debated whether shadcn’s copy-paste ownership model is better than traditional packaged UI libraries, with some seeing it as practical vendoring and others as frontend churn. (fatigue with AI-sounding product announcements, Base UI versus Radix as a default choice, copy-paste component ownership versus versioned UI libraries)
0:00 / 0:25software A free compilers textbook resurfaces, and HN debates the “language design” label↻ from 2021
This 2021 free online textbook resurfaced on HN today: Douglas Thain’s “Introduction to Compilers and Language Design,” developed for Notre Dame’s CSE 40243 compilers course. The book provides chapter PDFs and supporting code resources for building a simple compiler for a C-like language that emits working X86 or ARM assembly. It matters as a structured, practical path into a classic CS rite of passage, though HN commenters questioned whether the title overpromises on “language design.”
Discussion: Mixed — The discussion is broadly appreciative of a practical undergraduate compilers resource, especially from people who enjoy compiler projects or say they studied with Douglas Thain. The main pushback is that the book appears centered on building a C-like compiler, so several commenters argue it is more an intro to compilers than a serious treatment of programming-language design. (Praise for hands-on compiler construction, Debate over compiler implementation versus language design, Comparisons with Dragon Book, Tiger Book, TAPL, PLAI, and Crafting Interpreters)
0:00 / 0:21software A 2014 anti-ORM classic returns: learn SQL first↻ from 2014
This 2014 essay resurfaced on HN today, arguing that ORMs can augment SQL but should not replace it. The author, writing from experience with Postgres, SQLite, SQLAlchemy, and Hibernate, says ORMs often obscure inefficient queries, encourage overly broad object loading, complicate schema changes, and force developers to understand both SQL and the ORM’s translation layer. The practical takeaway is not “never use libraries,” but: treat the database and its queries as a real API, and learn SQL well enough to control what actually runs.
Discussion: Mixed — The thread is broadly sympathetic to the article’s core warning—ORMs do not remove the need to understand SQL—but not uniformly anti-ORM. Many commenters argue for a middle ground: use ORMs or query builders for boilerplate, type safety, migrations, and simple CRUD, then drop to SQL for complex or performance-sensitive queries. (SQL remains table stakes for serious database work, ORMs are useful for boilerplate, mapping, migrations, and conventions, Concern about hidden inefficient queries, N+1 behavior, and leaky abstractions)
0:00 / 0:35software Zig moves package management out of the compiler
Andrew Kelley says Zig has moved package-management functionality from the compiler into the build system’s maker process. That shifts package fetching, HTTP networking, TLS and crypto, Git protocol support, compression formats, and build.zig.zon handling out of the compiler binary and into source-shipped build-system code, making it easier to patch without rebuilding the compiler. The change is described as mostly non-breaking, shrinks a no-LLVM ReleaseSmall Zig executable from 14.1 to 13.5 MiB, and is tied to upcoming build-server work needed to unblock ZLS before Zig 0.17.0.
Discussion: Mixed — The thread was split between admiration for Zig’s visible craftsmanship and skepticism about why package-management logic lived in the compiler in the first place. Several commenters framed the change as healthy architectural cleanup, while others saw it as another example of young language ecosystems revisiting avoidable design choices. A side discussion focused on build-script sandboxing, including whether a future WebAssembly-based build system would be useful or merely security theater. (architectural cleanup versus initial design mistake, enthusiasm for Zig’s development culture, build-system sandboxing and security boundaries)
0:00 / 1:19security F-Droid blasts Google’s Android developer verification plan
F-Droid published a sharply worded attack on Google’s Android Developer Verification program, arguing that a system Google says is meant to fight malware will instead let Google block apps from developers who have not registered centrally. The post says rollout begins September 30 in Brazil, Indonesia, Singapore, and Thailand, with broader rollout expected later, and warns that F-Droid’s open-source distribution model may be incompatible with Google-controlled verification. The core issue is whether Android remains a user-controlled platform where sideloading is possible, or becomes a more centrally permissioned ecosystem.
Discussion: Negative — HN reaction is overwhelmingly hostile to Google’s Android Developer Verification, framing it as a loss of user control and a step toward centralized gatekeeping. The discussion is more mixed on F-Droid’s presentation: several commenters agree with the concern but think the “virus” and “malware vendor” rhetoric weakens the case. Much of the thread turns practical, with users debating whether GrapheneOS, mobile Linux, or other alternatives can realistically replace mainstream Android given banking, government ID, messaging, and app-compatibility constraints. (User ownership versus platform control, Fear that Google will become the sole arbiter of allowed Android software, Skepticism that developer verification meaningfully stops malware)
0:00 / 1:09security YouTube comments can prompt-inject Ask Studio
A researcher says YouTube Studio’s Ask Studio AI assistant can be influenced by malicious YouTube comments: a creator asks the assistant to summarize comments, and the model may treat a comment’s text as instructions. The proof of concept escalated from adding fake-looking “YouTube” notice text to generating a link that, if clicked, sent a private video title in the URL. The author says Google declined to treat it as a security bug because it required social engineering, while arguing the real issue is that YouTube’s own AI product is presenting attacker-controlled content inside a trusted creator tool.
Discussion: Mixed — HN largely treated the report as a real security concern and criticized Google/YouTube’s reported classification of it as social engineering. The discussion split over whether prompt injection is meaningfully fixable, whether requiring a creator to click a generated link lowers severity, and whether bug-bounty incentives discourage reporting. A few commenters tried to reproduce the issue and saw weaker or partially mitigated behavior, while others praised the writeup’s clarity. (Prompt injection in AI features that ingest user-generated content, Debate over social engineering versus product trust boundary violation, Skepticism about Google/YouTube vulnerability handling and incentives)
0:00 / 0:19security Anna’s Archive puts a $200K bounty on Google Books scans
This 2025 bounty resurfaced on HN today: Anna’s Archive is offering $200,000 for a scalable way to obtain Google Books scans, or similarly large rare-book collections, including material held by AI companies. The bounty page says Google Books exposes many scans only as search snippets, and explicitly invites anyone with access at Google to “sneak out” the data. That makes this less a normal archival project and more a flashpoint over data exfiltration, copyright, and whether preservation goals justify illegal access.
Discussion: Mixed — HN was split between strong support for shadow libraries as access and preservation tools, and concern for authors, copyright, and the legal risk of exfiltrating data. Several commenters described real barriers to buying books internationally, while others pushed back that book piracy can directly harm writers. A side thread also warned that one linked Anna’s Archive domain appeared to lead to suspicious or malicious behavior. (access to knowledge in restricted markets, copyright duration and author compensation, legal risk for Google insiders)
0:00 / 0:25security Claude Code Minecraft detour raises session-leak fears
A Claude Code GitHub issue reports an enterprise ZDR user seeing the agent suddenly talk about building a Minecraft temple, raising fears of possible session or cache leakage between workspaces or accounts. The reporter also noted an unusual setup involving a `.claude` context directory and work being done from another directory, and HN commenters pointed to a possible in-context trigger: a path containing `minecraft.py`. A commenter identifying themselves as from the Claude Code team said they were confident it was a hallucination but were looking into it, which matters because even the appearance of cross-session leakage can undermine trust in enterprise AI tooling.
Discussion: Mixed — HN was concerned but not convinced this proves a cross-account leak. Many commenters argued the Minecraft tangent could be a hallucination triggered by a `minecraft.py` path in a long context, while others treated any apparent response mix-up in an enterprise ZDR workspace as a serious security issue that needs a clear investigation. (session or cache isolation risk, LLM hallucination versus infrastructure bug, enterprise ZDR trust and compliance concerns)
0:00 / 0:34security Flipper Zero commits to maintained firmware after community pushback
Flipper Devices says it will continue maintaining Flipper Zero firmware and support community contributions, after backlash over the impression that official firmware development had effectively stopped. The new process moves feature requests to GitHub Discussions with voting, adds stricter pull-request rules, calls out AI-generated low-level code as requiring extra scrutiny, and makes integration and regression testing mandatory for firmware changes. The company says the device’s 700 KB firmware flash limit led it to move many functions into dynamically loaded microSD apps, and that firmware 1.0 in 2024 stabilized the API and SDK while the team shifted focus to new devices.
Discussion: Mixed — The HN thread was split and somewhat derailed. Some commenters saw the plan as reasonable for a mature device whose core firmware can be considered largely finished, while others read it as minimal life support or criticized the official project’s handling of pentesting features and alternate firmware discussion. A large early subthread focused on the article’s furry-themed artwork rather than the development plan. (Skepticism that official firmware is only being kept on minimal maintenance, Debate over whether mature software projects can be declared finished, Frustration from users who prefer unofficial firmware and broader pentesting features)
0:00 / 0:32hardware OpenPrinter promises a repairable printer, but HN wants proof
Open Tools is pitching OpenPrinter: a compact, repairable inkjet-style printer with refillable cartridges, CUPS-based network printing, sheet and roll-paper support, and parts intended to be standard, 3D-printable, or replaceable. The design uses region-specific HP cartridges, supports black-only or color-only printing, and is advertised as avoiding refill lockouts. The project also says files will be released once the final version is ready, under Creative Commons BY-NC-SA 4.0, which immediately raised questions about how “open” the device really is.
Discussion: Mixed — HN is broadly sympathetic to the goal of a repairable, refillable printer, but skeptical about execution and the meaning of “open” here. Commenters like the anti-DRM, repairability pitch, while questioning inkjet reliability, dependence on HP cartridges, crowdfunding risk, and the non-commercial Creative Commons license. (enthusiasm for refillable ink and avoiding printer DRM, skepticism about inkjet engineering and reliability, concern over reliance on HP cartridges and printheads)
0:00 / 1:18startups A startup parable about shipping the oven before it bakes
Weli.dev’s “Half-Baked Product” is a satirical startup story about a founder who raises money to build a smarter industrial oven, an engineer who wants to build the perfect version, and a sales motion that keeps promising features before the core product works reliably. The oven’s basic algorithm still burns bread and cakes, but the team gets pulled into enterprise customizations and novelty buttons because those promises help close deals and support fundraising projections. The piece lands because it turns familiar startup failure modes—oversized market slides, weak validation, sales-driven roadmaps, and mounting technical debt—into a simple product allegory.
Discussion: Positive — HN strongly liked the essay’s writing and found the startup dynamics painfully familiar. The mood was amused but uneasy: many commenters treated the oven story as an accurate allegory for VC-backed product drift, sales-led roadmaps, and unresolved core reliability problems. (Praise for the satire and storytelling, Founder ambition without domain expertise, Sales promises outrunning engineering reality)
0:00 / 0:23policy EU governments try to revive chat-scanning carve-out before recess
EU member states in the Council are trying to revive an expired exemption that let providers voluntarily scan private communications for child sexual abuse material and grooming signals, despite the EU’s e-privacy protections for communications confidentiality. Heise reports the Council is using a largely identical new regulation and an urgent procedure that could put pressure on the European Parliament just before its summer break. The stakes are both procedural and technical: supporters call the scans an essential child-protection tool, while critics see a privacy-invasive workaround and a way to normalize message monitoring while the broader Chat Control 2.0 fight remains stalled.
Discussion: Negative — HN’s reaction is overwhelmingly hostile to the Council’s maneuver, with users framing it as an end-run around privacy protections and parliamentary scrutiny. The main note of nuance is that several commenters stress this is Chat Control 1.0—a revived voluntary-scanning exemption—not the more sweeping Chat Control 2.0 proposal targeting end-to-end encrypted messengers. (Strong privacy and civil-liberties objections, Anger at fast-track timing before the summer break, Confusion between Chat Control 1.0 and 2.0)
0:00 / 0:30policy Gamers argue the real fight is ownership, not discs
A Bear Blog essay argues that the debate over physical versus digital console games misses the point: the real issue is whether players can own, transfer, preserve, and use games without depending on a single vendor. The author says discs still provide a practical path for resale, lending, rental, and archival dumping, while digital-only console ecosystems make users dependent on storefronts, licenses, and servers. The piece contrasts consoles with PC, where DRM-free stores and a more open platform can give players more options, though HN commenters pushed back that Steam still does not allow true resale or inheritance of licenses.
Discussion: Negative — HN was broadly alarmed about digital-only distribution and platform lock-in, with many commenters framing it as an anti-consumer loss of resale, lending, preservation, and long-term access. The main disagreements were over whether regulation is the answer, whether PC and Steam are really better, and whether physical media’s inconveniences weaken the case. (consumer ownership versus licensing, resale, lending, and transfer rights, game preservation and delisting risk)
0:00 / 1:11general Organic Maps gets a fresh HN spotlight as the offline, privacy-first map app
Organic Maps is a free, open-source, OpenStreetMap-powered navigation app focused on fully offline use for hiking, cycling, driving, and travel, with no ads, tracking, registration, or data collection claimed by the project. The site says the app supports offline search, downloaded regional maps, turn-by-turn navigation, CarPlay and Android Auto, contour lines, Wikipedia articles, and bookmark import/export. The HN discussion turned this from a simple app recommendation into a broader debate about the state of privacy-preserving maps, OpenStreetMap tooling, and governance after the CoMaps fork.
Discussion: Mixed — The thread is broadly enthusiastic about offline, open-source maps and OpenStreetMap-powered navigation, especially for hiking, travel, and low-connectivity situations. But the mood is tempered by repeated concerns about Organic Maps governance, the CoMaps fork, licensing of map data files, search quality, and whether it can really replace Google Maps for daily use. (Strong appreciation for offline maps, battery life, privacy, and no-tracking design, OpenStreetMap contributions and companion tools like StreetComplete drew praise, Several commenters pushed CoMaps as a fork after governance concerns around Organic Maps)
0:00 / 1:12general Costco’s low-tech logistics make it Amazon’s opposite
Phenomenal World argues that Costco is the “anti-Amazon”: instead of infinite assortment and fast home delivery, it uses a limited catalog, bulk in-person shopping, pallet-scale cross-docking, and very low overhead. The piece says that constraint is part of Costco’s value proposition: around 4,000 SKUs per warehouse, high membership renewal, fast-moving inventory, and a simpler logistics system that can support lower prices and relatively higher wages. It frames Amazon’s logistical sophistication as impressive but socially costly for everyday goods, while still acknowledging that fast delivery can be genuinely valuable for things like medicine and accessibility.
Discussion: Mixed — HN was broadly appreciative of Costco’s model, especially its curated selection and problem-avoiding simplicity, but commenters heavily challenged the article’s last-mile logistics argument. The main split was contextual: Costco looks elegant for suburban bulk shopping, while Amazon or delivery services may be more efficient for dense cities, remote workers, specialty items, or people without cars. (admiration for Costco’s constrained SKU model and buyer curation, skepticism that in-person bulk shopping is always more efficient than delivery, urban versus suburban retail logistics)
0:00 / 0:31general The case for learning something, even badly
Marginalia’s essay argues that adults can still learn practical skills—music, languages, drawing, typing, modeling, crafts—if they make space for short, consistent practice. Its core point is expectation-setting: early practice often feels bad, progress may happen between sessions and during sleep, and the goal is to get through the painful beginner phase into useful mediocrity. The piece matters because it frames learning less as a productivity hack and more as a long-term way to build agency and a richer life.
Discussion: Mixed — HN largely liked the essay’s push toward deliberate, low-stakes learning, but the discussion quickly complicated the simple “replace scrolling with practice” framing. Commenters emphasized mental energy, anxiety, parenting interruptions, and the difference between consuming tutorials and actually practicing; several also pushed back on the idea that AI or translation tools make personal knowledge obsolete. (Time versus energy and uninterrupted attention, Learning by making mistakes, not just consuming material, Anxiety, burnout, and psychological readiness)