HackerNews Digest

NIST researchers have demonstrated a monolithic 3‑D photonic integration platform that can generate arbitrary wavelengths on a single silicon wafer. The process stacks silicon dioxide, lithium niobate, and tantalum pentoxide (tantala) layers, using metal electrodes to control nonlinear frequency conversion and on‑chip switching. Tantala’s broadband nonlinear response converts an input laser line into a full spectrum spanning visible and infrared wavelengths, while lithium niobate provides fast electro‑optic modulation. Approximately 10 000 photonic circuits—each yielding a distinct output color—were fabricated on 50 fingertip‑sized chips mounted on a coaster‑sized wafer. The technology enables compact, low‑power laser sources across the spectrum, addressing the current limitation of fixed‑wavelength semiconductor lasers. Potential applications include quantum devices (optical atomic clocks, quantum computers) that require atom‑specific laser lines, AI hardware interconnects, and advanced VR displays. The work, detailed in Nature (April 15 2026, DOI 10.1038/s41586-026-10379-w), establishes a scalable route for integrated photonics, with collaboration from Octave Photonics for future up‑scaling.

The page hosts an open‑source “Tokenomics – Anthropic Token Cost Calculator” that aggregates community‑submitted data on token usage for Anthropic’s Opus models. It presents anonymous request‑token comparisons, highlighting performance differences between Opus 4.6 and Opus 4.7 on real‑world inputs. The dataset consists of stored rows identified only by anonymous submission IDs, ensuring no personal attribution. The tool is not affiliated with or endorsed by Anthropic.

Gun Rocket, originally created about a decade ago, was developed in Unity 4.6.0p1 (2015) and later migrated to Unity 5.5.0f3 (2018). The project’s version file confirms the latest editor version as 5.5.0f3, though Unity’s versioning scheme has shifted from simple numbers to year-based and now to high‑range numeric identifiers (e.g., 6000.4.1f1). Unity Hub lists only recent releases and pre‑releases, requiring access to the archive for older versions; the archive includes Unity 5 but not 4.6. Attempts to open the project in Unity 5.5.0f3 or 5.6.7f1 result in immediate closure with no log output, likely due to legacy license checks predating Unity Hub. The author plans to update the game for 2026, necessitating retrieval of a compatible Unity editor from the archive and addressing potential licensing or compatibility issues before proceeding with migration.

German instructor Grit Matthias Phelps at Cornell University introduced a semester‑long “analog” assignment in spring 2023, requiring students to complete a German‑language writing task on manual typewriters. The change responded to rising AI‑generated work and aimed to restore a technology‑free writing process, eliminating spellcheck, dictionaries, delete keys and screen distractions. Students received thrift‑shop typewriters with German or QWERTY layouts and were taught basic mechanics—paper loading, carriage return, and the line‑ending “ding.” The exercise forced slower, more deliberate composition; without instant corrections, learners relied on peer discussion and pencil marks for revisions. Participants reported heightened focus, increased interaction, and physical challenges such as weak pinky muscles and, for one student, a broken wrist. The initiative reflects a wider academic shift toward pen‑and‑paper exams and oral assessments to curb AI misuse, emphasizing manual skill development and reflective thinking.

The Astro Compass on early B‑52 bombers combined an optical Astro Tracker with an electromechanical “Angle Computer” to perform automated celestial navigation before digital computers and GPS. The tracker used a photomultiplier‑tube telescope on a stabilized platform to lock onto a star; its azimuth and altitude were fed electrically via synchro transmitters. Navigators entered star data (Sidereal Hour Angle, declination) and aircraft position (latitude, longitude, time) from the Air Almanac using the Master Control Panel, which displayed three selectable star sets. The Angle Computer mechanically modeled the celestial sphere on a 2 ⅝‑in half‑sphere. Input gears set the declination arm, the Local Hour Angle arm, and the latitude arm; a star pointer moved on the sphere, driving a semicircular azimuth arc and a slider. The arc’s rotation gave azimuth, while the slider’s teeth generated altitude, both output as synchro electrical signals. Internally, motors, differential gear trains, and feedback‑controlled servo amplifiers (vacuum tubes and transistors) positioned the mechanism and produced the outputs. By solving the spherical “navigational triangle” (using SHA, declination, latitude, and LHA), the system supplied a heading accurate to 0.1°, and its Line‑of‑Position display enabled position fixes from multiple stars. The electromechanical approach was chosen over resolvers or digital computers due to size, power, and reliability constraints of the early‑1960s.

Zero‑Copy GPU Inference from WebAssembly on Apple Silicon demonstrates that a Wasm module’s linear memory can be shared directly with the Apple‑silicon GPU, eliminating copies and serialization. The pipeline consists of three verified links: (1) `mmap` allocates page‑aligned, pinned memory (16 KB alignment) required by Metal; (2) Metal’s `makeBuffer(bytesNoCopy:length:)` wraps the same pointer as a GPU buffer, confirming pointer identity and showing negligible RSS increase (≈0.03 MB vs 16.78 MB for an explicit copy); (3) Wasmtime’s `MemoryCreator` supplies the mmap region as the module’s linear memory, exposing the identical pointer via `memory.data_ptr()`. A 128 × 128 GEMM test validates zero‑copy correctness and identical latency (~6.75 ms) to the copy path. Integrating this primitive with Apple’s MLX framework, a Rust‑compiled transformer (Llama 3.2 1B, 4‑bit, 695 MB) runs on an M1 MacBook Pro with per‑token generation ≈9 ms and negligible Wasm‑GPU dispatch overhead. KV‑cache tensors are serialized to safetensors (≈66 KB/token) and restored in ≈1.4 ms, yielding a 5.45× speedup over recomputation for 24 tokens; the benefit scales linearly with context length. The author’s “Driftwood” runtime will build on this zero‑copy foundation to enable stateful Wasm actors, snapshot mobility, checkpoint portability, and multi‑model support.

Japan accounts for 28 % of passenger‑kilometres, far exceeding other developed nations. The system is fragmented among many private firms and six regional JR companies that originated from the former state‑run JNR, which was privatized in 1988. Private railways generate most profit from non‑transport side businesses—real‑estate, retail, hotels, hospitals, and entertainment—built around stations, creating a virtuous “railway‑as‑city‑builder” model supported by liberal land‑readjustment rules that enable dense, transit‑oriented development. National zoning is permissive, allowing private developers to acquire and redevelop land near lines, while 30 % of urban land has undergone readjustment projects. Parking is privatized, requiring owners to secure spaces before buying cars, which limits car use and internalizes its external costs. The government imposes modest fare caps and provides capital‑expansion subsidies tied to public goals (e.g., accessibility, earthquake‑proofing) but does not subsidize day‑to‑day operations. Combined vertical integration, productive labor reforms after JNR privatization, and targeted subsidies have produced a profitable, highly utilized railway network that can be emulated elsewhere.

The post examines how reducing CI worker setup time can dramatically improve overall build cost, especially with Intercom’s 1,350‑worker default. The author focuses on speeding Ruby boot time by optimizing Bootsnap, a load‑path cache used in Rails. Bootsnap replaces Ruby’s linear `$LOAD_PATH` search (O(N·M)) with a hash lookup built by scanning all load‑path directories once, but its cache must be invalidated when files change, requiring many `stat(2)` calls. On large monorepos this scan costs ~1 s, prompting a redesign to eliminate N+1 syscall overhead. By extending `Dir.foreach` (and later proposing a new `Dir.scan` method) to expose file‑type information from `readdir(3)`’s `d_type`, the scanner avoids per‑entry `File.directory?` calls, cutting the scan from ~500 ms to ~230 ms—a 2.16× speedup, enabling scanning of ~32 k files across ~10 k repos in ~230 ms. The author also notes secondary gains from optimizing `File.join`. The proposed `Dir.scan` API is slated for Ruby 4.1.0.

- Gustave Eiffel’s original Eiffel Tower design (1889) included a private office above the top floor, accessed via a narrow spiral staircase (1,062 steps, single guardrail) linking the second and third levels. - The staircase was removed in 1983 when new elevators were installed and divided into 24 sections; pieces are displayed at various locations, including the Musée d’Orsay and the Statue of Liberty. - One restored steel‑and‑riveted‑sheet‑metal fragment, about nine feet tall and containing 14 steps, is being auctioned by Artcurial on May 21, with an estimated price of $141 k–$176 k. The most expensive previous fragment sold for €523,800 in 2016. - The Eiffel Tower, 1,024 ft tall, was built for the 1889 World’s Fair, held the record for tallest structure until 1929, and was originally slated for removal after 20 years but was retained for scientific use. - Today the tower receives ~7 million visitors annually (≈300 million total since opening) and featured in the 2024 Paris Olympics, where medals incorporated recycled tower iron.

Modern Common Lisp with FSet – Version 1.0 for FSet v2.4.2, © 2026 Scott L. Burson. The document is released under the Creative Commons CC BY‑NC‑SA 4.0 license, which permits non‑commercial redistribution, remixing, adaptation, and distribution in any medium or format provided that (1) proper attribution to the author is given, (2) the work is not used for commercial purposes, and (3) any derivative material is licensed under identical terms. The author explicitly states that the text contains no LLM‑generated content and that all punctuation, including em‑dashes and semicolons, was manually written. Readers are invited to report errors or suggestions by filing issues on the Common‑Lisp.Net GitLab instance or on GitHub.