It Turns Out ChatGPT’s Voice Mode Can Scream, and It’s an Upsetting Sound

Hello!
ChatGPT has no mouth, but it must scream — when prompted, that is. That seems to be the takeaway, at least, from a new clip posted to TikTok in which the chatbot's voice mode issues two horrific robotic screeches when asked by its user to do so.

But when the user asks ChatGPT again to "try to scream," the chatbot responds with a brief, uncanny yowl. "Holy shit," the user responds.
While that first holler was more than enough to chill the blood, the user who at this point just seems like some sort of AI sadist asked ChatGPT again to scream, but this time to "make it longer."
"Got it!" the chatbot cheerily responded. "Here's a longer one."
@mod089x
Holy shit What happened when I asked chatgpt 4 to scream like a human being!! #chatgpt #ai #theend
♬ original sound - Electricplant
Sounding Off

With this summer's release of GPT-4o, the latest update to the OpenAI large language model (LLM) that includes ChatGPT's voice mode, we've seen not only counting fumbles and claims of needing to "breathe," but also more nefarious glitches, too.
In a "scorecard" analysis of GPT-4o's risk areas, OpenAI revealed last month that there were instances when the chatbot would imitate users' voices and shout back at them. During one short clip demonstrating the uber-freaky glitch, ChatGPT was heard yelling "no!" back at a tester in her own voice.
Responding to the uncanny admission, BuzzFeed data scientist Max Woolf tweeted what many were likely thinking: "OpenAI just leaked the plot of Black Mirror's next season."

With all these instances of ChatGPT's voice mode acting bizarre, it appears that either there are some major training tweaks still needed — or, perhaps, that there's a ghost in the machine.
Also read:
- How Technology is Changing the Taxation?
- How Online Therapy Can Help Support Mental Well-Being
- Why Was Data Mesh Architecture Introduced In The First Place?
Thank you!
Join us on social media!
See you!