HackerNews Digest

The Authors Guild and a group of prominent writers have filed a class‑action suit against OpenAI and Microsoft, alleging that the companies trained their AI models on copyrighted books obtained from the LibGen repository without permission. Recent unsealed briefs cite internal communications showing executives were aware that such “mass piracy” was illegal and would displace human authors. Key points include:

• OpenAI policy director Jack Clark warned in 2020 that GPT development would increasingly substitute for writers and cause unemployment.
• Former OpenAI researcher Tarun Gogineni described his goal of having GPT write the final two books of Game of Thrones and dismissed author complaints as “acceptable economic disruption.”
• Microsoft learned of OpenAI’s LibGen use as early as April 2019 during a presentation to Bill Gates and CTO Kevin Scott.
• Internal messages reveal concerns focused on “optics” rather than legal risk, and a “Project Clear” effort in mid‑2022 to delete LibGen files from OpenAI’s storage.

The case is proceeding as multidistrict litigation in Manhattan, with a hearing anticipated in early 2027. The Authors Guild, representing over 18,000 writers, is leading the lawsuit.

The post revisits a 2021 review of Henry George’s Progress and Poverty and updates five‑year‑old arguments about land‑value tax (LVT). It notes rapid legislative momentum: Virginia and Kentucky passed LVT‑enablement laws in 2026, with pending bills in Washington, New York, Ohio, and international interest from the UK, South Korea, and Germany. The author argues that land remains a dominant source of wealth, especially in dense urban cores and in South Korea where land‑to‑GDP ratios exceed 500 %. Empirical work using open‑source tools (CivicMapper, LVTShift) shows most LVT gains accrue to median homeowners, while vacant land and surface parking lose value. Research on tax incidence confirms LVT is rarely passed to tenants except under pervasive rent control with special exemptions, as observed in Denmark. Assessment challenges are deemed manageable: modern appraisal systems can separate land from improvements, though accuracy varies. The piece emphasizes practical “scut work”—local lobbying, data‑driven models, and case‑study playbooks—as the primary driver of policy adoption.

A FLIP fluid simulation was implemented on a 13×28 electromechanical flip‑dot panel for the EMF2026 installation. The author acquired surplus Hanover panels (2007) from “Look Mum No Computer”, noting the board overhang prevents seamless tiling and the original driver updates the whole matrix in ~1 s. Each dot flips in ~60 ms with a ≤1 ms polarity pulse; columns can be driven in two passes (high then low) to potentially achieve ~56 ms refresh if pulses are ~1 ms. High current (~0.7 A at 12 V, up to >1 A for faster pulses) and simultaneous coil activation present power and thermal challenges. Desoldering dots proved impractical, so the solution uses inexpensive motor‑driver H‑bridge ICs (MX6208/BE6208/LK6208, SOIC‑8, 0.5 A continuous) mounted on a custom board that replaces column traces. Each 8‑bit shift register (74 595) drives four dots directly; by tying one H‑bridge input high, a single‑ended operation yields push‑pull output without extra diodes. Capacitor‑coupled half‑bridge alternatives were explored but rejected due to size, cost, and quiescent current of BJTs. The final PCB design incorporates 15.24 mm pitch flip‑dot footprints, keep‑outs for test pads, and uses KiCad despite its limited repetitive‑component workflow.

The mini‑book “Go concurrency distilled” is a concise refresher on Go’s concurrency primitives, illustrated with runnable examples. It covers:

• Goroutines – lightweight functions started with go, managed by the runtime, and synchronized with sync.WaitGroup (including a helper WaitGroup.Go).
• Channels – synchronous communication (ch <- v / v := <-ch), with patterns for output channels, closing, iteration (range), directional types, buffered channels (make(chan T, n)), and nil channels.
• Select – channel‑focused multiplexing that chooses ready cases, supports cancellation, non‑blocking sends, and default handling.
• Pipelines – composition of stages where each step reads from an input channel and writes to an output channel; includes output, done, and cancel channels for signaling completion or early termination.
• Error handling – strategies for propagating errors in pipelines: early return on first error, or wrapping results in a struct containing both value and error.

Additional topics listed (time, context, wait groups, data races, mutexes, semaphores, atomics, testing, scheduling, diagnostics) indicate further coverage, though the excerpt focuses on the core concurrency mechanisms.

PipePipe is an open‑source Android application forked from NewPipe in early 2022, operating as an independent project that does not track NewPipe updates. It offers a re‑imagined YouTube client with numerous enhancements: SponsorBlock integration for skipping ads, ReturnYouTubeDislike restoration, original (non‑localized) titles, optional login for restricted or premium content, live‑chat overlay in danmaku style, AV1 and VP9 codec support, background music playback, advanced search filters, keyword/channel filters, Shorts and paid‑video blocking, swipe‑to‑seek and fullscreen gestures, long‑press speed control, sleep timer, batch playlist downloading, and searchable local playlists/history. Cookie handling is limited to playback stream retrieval when a login cookie is set. Development focuses on rapid issue resolution and frequent feature updates; contributions via issues and pull requests are welcomed, though service requests are not accepted. Community contributors include @Priveete for SABR support and AioiLight for NicoNico code. Funding options are listed on Liberapay and Ko‑fi.

DeepSeek Elastic Compute (DSec) is presented as a sandbox infrastructure designed to support large‑scale agentic training. The system emphasizes elasticity and sandboxing to manage compute resources efficiently while maintaining isolation for training agents. The accompanying figures reference external resources and affiliations—archive, BibSonomy, Reddit, Simons Foundation, and Schmidt Sciences—suggesting data sources, citation repositories, or institutional collaborators involved in the work. No additional methodological or experimental details are provided in the extracted text.

The Primer design system’s CSS‑in‑JS approach began causing performance regressions as component counts grew in 2023, leading to longer client‑side style initialization, slower server‑side rendering (SSR), and unwieldy style updates. The team replaced it with CSS Modules, authoring component‑scoped CSS files that compile into static stylesheets, eliminating runtime CSS processing. Migration proceeded incrementally: each component received a translated CSS Module file, was gated behind a feature flag, and verified against visual regression snapshots before gradual rollout. By Dec 2024 all Primer components used CSS Modules, achieving ≈55 % faster SSR and ≈25 % quicker component initialization. Subsequent work removed the “sx” prop and styled‑components across GitHub. A VS Code plugin and internal codemods migrated ~7,760 sx props (April 2025–May 2026), reducing them to zero by June 2026; SSR time improvements of 1–22 % were observed on affected pages. The final step decoupled theming from styled‑components, completing a full transition to 100 % CSS Modules by June 2026.

reladraw is a lightweight, TypeScript‑based diagram description language that lets users specify relative positions of nodes instead of absolute coordinates. It bridges the gap between declarative tools such as Mermaid, Graphviz, or D2—which auto‑place elements—and manual editors like draw.io or Excalidraw, which require explicit placement. The syntax defines nodes (e.g., node app "Web app") and edges with directional hints (edge app.api -> store "queries" from: right to: left). Positions are expressed relative to other nodes (below app.ui, right of app level with app). The repository provides a CLI (reladraw file.reladraw -o file.svg) and npm installation (npm install -g reladraw). A skill package can be added to various coding agents via npx skills add reladraw/reladraw. Version 0.8.0 is an early release with a parser, layout engine, and SVG renderer that have no runtime dependencies. The project is Apache‑2.0 licensed (code only) and invites issues—especially diagrams that cannot yet be expressed—as contributions.

C’s native integer types (char, short, int, long, long long) were intentionally sized to match each machine’s “natural” word, providing minimum guaranteed ranges rather than fixed widths. In the 1970s hardware varied widely—bytes could be 6‑9 bits, word sizes spanned 12‑64 bits, and signed representations included two’s‑complement, ones‑complement, and sign‑magnitude. Fixed‑size types would have forced costly multi‑instruction arithmetic on 16‑bit CPUs, extra masking on 36‑bit machines, and software emulation on non‑two’s‑complement hardware. The C standard codified this philosophy with minimum ranges and, from C99, optional exact‑width types (e.g., int8_t) plus required least/fast types (int_least32_t, int_fast32_t). Projects such as Lua demonstrate portable code that queries ranges (e.g., using UINT_MAX) and selects types accordingly, avoiding assumptions about byte size (CHAR_BIT ≥ 8). Modern languages often fix widths because contemporary platforms have converged on 8‑bit bytes and two’s‑complement arithmetic, yet they retain a natural‑size integer (e.g., Rust’s isize, Go’s int). Thus, C’s flexible integer sizes were a deliberate design for portability, not a mistake.