On: Whisper – open source speech recognition by OpenAI

[datalopers]

If you know which 2-3% are the false positives, you have a very lucrative business model.

[MonkeyMalarky → datalopers]

When doing validation, I find it will often be the same errors repeated again and again in a transcription. Like it will fail on someone or some thing's name (that is rare / unique) and map it onto a known similar sounding word.