`April`: voice assistant for email and calendar on the go

// related

• #0001

  #0001Idea signalsGeekNews6 hours ago

  `AI Overviews` Can Be Skewed by a Single Page

  40radar

  A lone webpage changed answers from ChatGPT and Google in a BBC test. Brand and reputation monitoring now needs answer-engine checks, not just classic SEO.

  • —BBC’s test made ChatGPT and Google identify one person as a world champion hot dog eater. One page was enough to distort the answer.
  • —Sensitive verticals like health and finance are exposed first. Thin but authoritative-looking pages can shift AI summaries.
  • —This creates a service angle: monitor AI Overviews and chatbot answers for brand, founder, product, and competitor queries.

  Source: news.hada.io/topic?id=29717Read original →
• #0002

  #0002Idea signalsHacker News · MRR7 hours ago

  `Phosphene` open-sources custom macOS video wallpapers

  50radar

  PhosphenemacOS wallpaper utility — adds custom videos to Settings

  Custom macOS video wallpapers now have an open implementation. Polished incumbents make this better as reusable reverse-engineering than a standalone business.

  • —WallpaperExtensionKit.framework controls how wallpapers appear in macOS Settings; Phosphene replicates enough behavior to list custom entries.
  • —The app keeps video running on the desktop, unlike Apple Aerials, using AVSampleBufferDisplayLayer with PTS-offset gapless looping.
  • —Thermal state, battery, brightness, and window occlusion drive pause or downshift logic. That is the practical part to copy.
  • —The commercial read is blunt: even a working niche utility loses appeal when existing competitors are already polished.

  Source: github.com/kageroumado/phospheneRead original →
• #0003

  #0003Idea signalsSimon Willison13 hours ago

  `TokenSpeed` makes LLM output speed visible

  40radar

  TokenSpeedHTML demo app — visualizes LLM token output speed

  Speed claims become easier to judge when they animate like real model output. Useful for product demos and latency expectations, not model ranking.

  • —Simulates 5 to 800 tokens/second, enough to compare slow local runs, normal API streams, and fast batch-like output.
  • —A quoted 30 tokens/second number is hard to feel in specs; visual playback makes perceived wait time obvious.
  • —The app is plain HTML with source available, so the idea is easy to embed in docs, sales pages, or model-picker tools.

  Source: simonwillison.net/2026/May/20/tokens-per-second/#atom-evRead original →