OpenAI's Whisper automatic transcription system, built on large language modeling technology, will sometimes output random strings that weren't in the audio input at all because the architecture loosely couples input to output, letting the language model do text fabrication, and all biases get amplified as a result.
factualpending
Speaker
Dr. Emily M. BenderEvidence Quote
“Whisper...will sometimes output just random strings that weren't in the audio input at all...because the architecture has changed from really being tightly coupled input to output to effectively letting the language model do its text fabrication thing.”
Source
Authors of "THE AI CON" discussing the problematic hype about AI & what those working tech should do— Product Impact Reports | AI Strategy & PlaybooksCreated: 8/11/2026, 7:30:16 AM
Your notes
Loading notes...