ChatGPT Struggles With Precise Timing; Sam Altman Says Voice Model Fix Is at Least a Year Away

OpenAI’s ChatGPT voice model can hold natural conversations but still struggles to track seconds as accurately as a conventional timer. That reflects how generative AI produces responses based on language probabilities rather than continuously consulting a reliable clock. Fluent answers therefore do not guarantee accuracy in tasks involving time, numbers or calculations, underscoring why users should not treat the model as a direct substitute for purpose-built tools.

A TikTok user recently asked ChatGPT to act as a timer and found that the model fabricated the number of seconds elapsed instead of counting down in real time. OpenAI CEO Sam Altman subsequently acknowledged the problem and said improvements to this type of timing capability in voice models would take at least about another year. OpenAI has yet to announce an exact launch date.