HackerNews Digest

Starting in May 2026, YouTube will add internal signals that automatically detect significant photorealistic AI usage in videos. If a creator does not manually disclose AI involvement, the system will apply an AI‑generated label to the content. Creators can contest a label in YouTube Studio, but disclosures will remain permanent for: • content created with YouTube‑provided AI tools such as Veo or Dream Screen, and • content that includes C2PA metadata indicating it is fully generative AI. The rollout aims to make AI disclosures more seamless while keeping creators able to adjust incorrect detections. Images in the post illustrate the interface for AI labeling, a simplified overview of the labeling and auto‑detection process, and the disclosure options presented at upload.

The author argues that advances in artificial intelligence will dramatically increase white‑collar productivity—potentially tenfold—across tasks such as work, learning, and social interaction. They propose that this productivity boost should allow a reduced workweek, specifically taking Friday off while maintaining output levels. The suggestion includes using AI agents to continue processing prompts created on Thursday, thereby minimizing the impact of the day off on overall results. The author extends the proposal to executives and board members, implying they could also benefit from an extra leisure day without sacrificing performance. Additionally, the writer references personal financial pressures, such as high childcare costs in California, to illustrate the practical motivation for a shorter work schedule.

No additional content was provided beyond the title, so a summary cannot be generated.

Anthropic and OpenAI have shifted to enterprise‑oriented pricing that mirrors API token costs, ending earlier “seat‑based” discounts. Anthropic’s Enterprise plan moved to $20 per seat plus usage in November 2025; OpenAI aligned Codex pricing to token usage in April 2026 for all plans. Heavy users can now spend roughly $1,200 per month per vendor on Claude Code and Codex, while paying only $100–$200 for subscription tiers. The companies’ focus on coding/general‑purpose agents (Claude Code/Cowork and Codex) appears to have reached product‑market fit, driving substantial enterprise spend that outpaces consumer‑subscriber revenue. Job listings show a large proportion of OpenAI (≈33 %) and Anthropic (≈27 %) openings in enterprise sales and support. Reported budget overruns at Uber and Microsoft’s license cuts illustrate rapid adoption and high usage costs. Anthropic has signed a $1.25 billion‑per‑month cloud‑compute contract with SpaceX, indicating massive inference budgets. API revenue is being eclipsed by direct enterprise contracts, with Anthropic rumored to approach $11 billion quarterly revenue and potential profitability. April 2026 is marked as a new revenue inflection point ahead of upcoming IPO filings.

Apple’s APNs (since 2009) and Google’s cloud messaging services (2010‑2016) form the only delivery pipelines for mobile push. Over the past fifteen years the platforms have moved from passive relays to active editors: Android 8 introduced developer‑defined notification channels with user‑controlled importance; iOS 15 added Focus, Scheduled Summary and a four‑level interruption taxonomy; Android 13 made POST_NOTIFICATIONS a runtime permission, causing opt‑in rates to drop from ~85 % to ~67 % (Batch 2025). Both OSes now run on‑device AI that can summarise, reorder, or suppress alerts. Apple uses a 3‑billion‑parameter foundation model with lightweight LoRA adapters (Apple Intelligence); Google’s Gemini Nano operates via AICore with similar adapter‑based specialization. These models decide visibility, ranking, and summarisation after the payload reaches the device, but developers receive no API to detect or influence the outcome. Compared with email, push lacks open‑rate metrics, authentication standards, and persistent storage, leaving senders with minimal instrumentation and increasing platform opacity. User studies show most notifications are low‑value, with messaging ranked highest and promotional alerts lowest, reinforcing the platforms’ defensive stance on interruptive content.

The post explains why pasting images (Ctrl + V) into Claude Code when it runs inside WSL via Windows Terminal fails and provides a workaround. **Broken pieces** 1. WSLg’s clipboard bridge only transfers Windows images to Linux as an old BMP variant (BI_BITFIELDS). Claude Code’s image parser cannot read it, so the paste silently fails. 2. After a custom script places a PNG on the Linux clipboard (e.g., with `wl-copy --type image/png`), WSLg syncs the Linux clipboard back to Windows, then pushes the Windows side back to Linux as the unsupported BMP, overwriting the PNG. The overwrite occurs without a notification to the watcher. 3. Windows Terminal intercepts Ctrl + V for its own text‑paste operation, preventing the keystroke from reaching Claude Code. **Fix** – three components: * `clip-listener.exe` (Windows) captures clipboard images and converts them to PNG using GDI+. * `wsl-clip-bridge` (WSL) runs `wl-copy` to set the PNG on the Linux clipboard and re‑asserts it after 0.5 s to recover from the BMP overwrite. * An Alt + V keybinding in `~/.claude/keybindings.json` triggers Claude Code’s `chat:imagePaste` handler, bypassing Windows Terminal’s Ctrl + V. Installation script compiles the listener, installs binaries, adds the bridge to `~/.bashrc`, and creates the keybinding. The bridge works until upstream fixes add PNG support to WSLg, BMP support to Claude Code, or Windows Terminal stops hijacking Ctrl + V.

SimCity 3000 can be run at native 4K on Windows 10 with the following fixes: - Replace the original executable with the GOG‑provided patched EXE (MD5 638eb5b3…) to add widescreen support and remove the CD check. - Edit **SC3U.ini** to set `ScrollMarginFactor=0.005787` and correct mouse acceleration. - Install the D3D9 wrapper (MD5 e97d0989…) in the **Apps** folder; edit **dxwrapper.ini** to set `[d3d9]EnableWindowMode=0` and `[FullScreen]FullScreen=1` for true fullscreen at 3840×2160. - Apply the 4 GB RAM patch (NTCore) to the SC3K.exe so the game can use more memory on 64‑bit systems. - Disable the dead‑online auto‑updater by overwriting **Apps/Updater/UpdateSettings.ini** with the provided patched file (MD5 e01ec9e7…). - Restore missing tracks by installing the music patch (MD5 5877e363…) and copying the original *.xa* files from **APPS/RES/SOUND/MUSIC** on the CD to the HDD location. These steps resolve the original low‑resolution limits, laggy scrolling, tile‑load artifacts, and missing audio, and have been confirmed on Windows 10 (Windows 11 untested).

The piece outlines how senior leaders should treat “infrequent but important” meetings by assuming each has a purpose and a specific request—“the Ask”—directed at them. Key guidelines: - Verify that the meeting exists for a reason and that your presence is expected to fulfill an ask. - In one‑on‑ones, focus on high‑signal, predictable topics; in ad‑hoc meetings, quickly surface the ask and offer help rather than guessing solutions. - For promotion‑related talks, ask “How can I help?” to uncover underlying requests such as desire for a manager role, compensation, or process clarification. - In cross‑team meetings where the agenda is unclear, explain your team’s planning cycle, allocation process, and past collaboration patterns; if the ask isn’t clear, schedule a follow‑up rather than commit. - Maintain long‑term relationships (“shared‑fate” meetings) even when immediate work isn’t apparent, as future alignment can prevent duplicated effort. Overall, senior leaders should rely on intuition built from repeated experience, surface the ask early, and use structured follow‑ups to align strategy without over‑committing.

The article examines three LoRa‑based mesh networking solutions—Meshtastic, MeshCore, and Reticulum—and argues that Reticulum is the most scalable and versatile. Meshtastic, an early consumer‑focused product, works well for small groups but relies on flooding, limited hop counts, and cannot scale to regional or larger public meshes. MeshCore introduces true routing and up to 64 hops, yet separates devices into “companions” and “repeaters,” requires companions to stay within repeater range, and depends on proprietary client software with pay‑walls, limiting openness and resilience. Both systems are tightly bound to LoRa, making inter‑technology integration (e.g., Wi‑Fi, microwave, fiber, internet, Tor/I2P) cumbersome and often reliant on hacky MQTT bridges. Reticulum provides a protocol‑agnostic routing stack that automatically selects optimal paths across heterogeneous links, supports global address space without central authority, and offers native applications (NomadNet, Sideband, Meshchat) using common protocols (LXMF, LXST, RRC). Its main drawback is the lack of standalone LoRa firmware; current RNode firmware requires an attached computer, raising power and cost for unattended nodes, though microReticulum ports aim to address this. The author concludes that Reticulum best meets the needs of large, interoperable public meshes.

The author jail‑broken a 7th‑gen Kindle Paperwhite and set up a Rust development workflow for it. Using cargo‑zigbuild with the Zig compiler (which bundles musl libc), the code is cross‑compiled for the Kindle’s ARMv7‑eabihf architecture via `cargo zigbuild --release --target armv7-unknown-linux-musleabihf`. SSH access is obtained through the USBNetwork tool; the public key is manually added to `/mnt/us/usbnet/etc/authorized_keys` because `ssh-copy-id` fails. A “Hello World” binary runs correctly, then the Slint GUI library is integrated. A custom `LineBufferProvider` rasterizes Slint’s output, writes grayscale frames to `/dev/fb0` (memory‑mapped), and triggers e‑ink refresh with an `ioctl` call using the dirty region supplied by Slint. Touch input is read directly from `/dev/input/event1`; the Kindle’s Linux multi‑touch B protocol events are accumulated (X, Y, tracking ID) and translated into Slint pointer events (`PointerPressed`, `PointerMoved`, `PointerReleased`) on each `SYNC_REPORT`. After extensive debugging of screen refresh and input handling, a functional backend crate was created and published on crates.io, enabling Slint‑based applications on the Kindle.