Whisper's major problem was hallucinations, how are the new models doing there? The performance of ChatGPT advanced voice in recognizing speech is, frankly, terrible. Are these models better than what's used there?