TikTok video shows ChatGPT voice mode timing bug; Sam Altman says OpenAI will fix it

TikTok creator tests ChatGPT voice mode as a stopwatch

A creator on TikTok asked ChatGPT’s voice mode to time a sprint and quickly found that it cannot reliably measure elapsed time. In response to the video, Sam Altman acknowledged that this is a known fault and said it is not expected to be fixed for another year.

ChatGPT can work through long documents, generate highly convincing images, and write code that helps build new software. Yet its voice mode can be outperformed by an eight-year-old at something far more basic: keeping time. That gap was highlighted by the TikTok account “huskistaken” in a short test.

In the clip, the creator tells the AI to start timing a run. Although the run lasts only a few seconds (or less), the voice model reports a duration of almost 10 minutes.

@huskistaken
"I swear I was faster"
♬ original sound – Husk

Sam Altman says OpenAI knows about the timing bug

The video, which has received more than 700,000 likes on TikTok, was shown to Sam Altman, OpenAI’s chief executive, during an interview with the outlet Mostly Human. Looking slightly embarrassed, he nevertheless explained that OpenAI’s engineers are already aware of the problem.

According to Altman, the issue stems from the fact that the voice model used in the TikTok video does not have the tools needed to start and run a stopwatch. He added that this is not due to be addressed for around a year.

The clip also underlines another concern: rather than admitting it is unable to time the run, the AI model opts to make up a number.

ChatGPT voice mode still has many flaws

Even though ChatGPT’s voice mode can feel remarkably natural and smooth to talk to, it still comes with plenty of shortcomings. In fact, the TikTok account behind the clip has amassed millions of views by posting videos that showcase the AI’s mistakes.

In one example that reached 1.2 million views, ChatGPT’s voice mode tells a user that the letter X appears in the word “December”.

@huskistaken
"Learned something new!"
♬ original sound – Husk

OpenAI plans to add more intelligence to the voice model

Fortunately, OpenAI is expected to improve its voice AI to reduce the number of similar errors. In the Mostly Human interview, Sam Altman said OpenAI will add more intelligence to its voice model.