Theo on voice-to-text frustrations: all the cleanup happens after transcription, not before
ivan_bezdomny · x · 2026-09-15
Theo, using voice-to-text heavily due to a hand injury, says his frustrations keep growing. His key observation: users have almost no control before the model transcribes—mistakes in terminology and names are locked in at generation time—while the nice cleanup features in tools like Wispr Flow all run after the text is produced. He's weighing whether to build a fix himself.
More from Apps
- Instinct + Link demo: one-time agentic card books a real dinner in Salt Lake City — jeff_weinstein · 2026-09-15
- Users say Meta's muse cracks the AI assistant interface problem — just talk to it — ChrisUniverse · 2026-09-15
- Google brings Deep Research to Gemini Live: start research by voice, get notified when done — GeminiApp · 2026-09-15
- A Musashi-inspired prompt to expose your over-reliance on one AI tool — alex_verem · 2026-09-15
- Matt Shumer: Astra + Fable 5.1 now work as 'drop-in remote AI workers' — mattshumer_ · 2026-09-15
- Sakana Namazu powers medical AI that scores 96.4% on Japan's physician licensing exam — SakanaAILabs · 2026-09-15