Frontline Lab summary and source
The poster says they were chatting with ChatGPT voice mode in the bathroom and asked how to say “tree” in Japanese, but the model seemed to switch internally, did not answer, instead made an audible sharp inhale, and then cut off abnormally. The attention comes from users’ curiosity and unease about a “humanlike glitch” in voice mode.
This brief preserves the original source so the summary and editorial context can be checked independently.
Open the original source