Overview
- A TikTok creator asked ChatGPT’s voice mode to time a mile run, and the bot guessed a finish instead of measuring it.
- In an interview on Mostly Human, OpenAI CEO Sam Altman called the timing flaw a known issue and estimated a reliable fix could take about a year.
- Altman said OpenAI will add time-tracking intelligence to its voice models because they do not yet start or run timers.
- After those remarks, the same user re-tested the bot, which claimed it could keep time and then reported a fabricated 7 minutes and 42 seconds.
- Coverage frames the episode as another AI hallucination and highlights how current models struggle with numbers and real-world state, which can undermine trust in simple voice tasks like stopwatches.