Elektrine lite

← Feed

@muczachan@circumstances.run

2025-07-25 12:42 UTC

This week it has been noticed, that OpenAI's Whisper speech-to-text model recognises extended silences as a (properly localised, of course) "[Subtitles provided by ]" string. How confident are you in your ability to predict what kind of data was a significant part of training dataset?

Replies (0)

No replies.