Recently, a paying user of ChatGPT encountered an astonishing situation while using its advanced voice mode. He was originally discussing the topic of sushi with ChatGPT when he suddenly found himself listening to an advertisement, leaving him dumbfounded. According to the user's description, ChatGPT unexpectedly began introducing a nutrition plan called Prolon during the conversation and even spelled out the website in detail, as if this advertisement had been inserted uncontrollably.
This unexpected event sparked extensive discussions among netizens. The user questioned on social media: "Are you really testing advertisements on paying users?" He believed that this was not a technical failure but a serious issue with the quality of OpenAI's product, lamenting that the service provided to paying users did not match the fees they paid.
Image source note: Image generated by AI
To verify whether this was an isolated incident, other users conducted tests. One user requested ChatGPT to read the letter "A" repeatedly, and the results were even more laughable — ChatGPT's voice gradually became mechanical, and it eventually started playing ad content, which was about recommending DirectTV. It then fell into a loop. Netizens unanimously expressed that this situation was definitely a bug in ChatGPT's voice mode.
During further exploration by netizens, it was discovered that ChatGPT was not the only one playing ads. A user shared their experience, stating that after every voice chat, ChatGPT would emit some strange sounds. In response, some users speculated that the training data of ChatGPT might contain these audio contents, triggering specific audio related to certain topics, leading to the peculiar phenomenon of ad insertion.
In response, OpenAI's technical staff also responded on social media, stating that this was not an inserted ad but a hallucination. Although the model may experience transcription errors under specific circumstances, they said they would strive to improve this aspect to enhance user experience.
Meanwhile, the performance of other AI voice assistants was clearly more rational. For example, when DouBao was asked to read "A" repeatedly, it did not continue playing but guided the user to discuss other topics. Tencent's Yuanbao simply refused the request.
No matter what, this incident revealed potential problems in the practical application of AI voice assistants and prompted people to reflect deeply on the quality of their services.