Cornell University researchers have identified concerning behavior in OpenAI's speech-to-text transcriber Whisper, revealing its potential to hallucinate violent language. The study sheds light on the ...