TikTok Fact-Checks ChatGPT About Its Timer Feature
OpenAI’s ChatGPT went viral on TikTok for claiming to have a feature it doesn’t. Sam Altman addressed the issue with the chatbot. When was ChatGPT forced to watch its creator admit that it doesn’t have the feature? ChatGPT still couldn’t admit its faults.
Husk, also known as @huskitaken on TikTok, creates content about his ChatGPT. He would often troll the bot and ask it random questions. But recently, a video of Husk asking ChatGPT to time his run went viral. In the clip, Husk only jogged for a few seconds, but the bot said he ran for ten minutes and twelve seconds. Essentially, the bot gave Husk a made-up response.
Altman, who watched the viral video, acknowledged that the ChatGPT model that Husk has doesn’t have the ability to start a timer. Even Altman had a good laugh about the situation and said that perhaps within a year, the issue could be resolved.
ChatGPT caught lying on camera
Husk went back to TikTok to create a strangely meta video. He made his ChatGPT react to Altman, who was reacting to the original, viral clip. Even after watching, the bot seemed averse to the concept of accountability.
Husk made his ChatGPT watch the clip, and the bot recognized Sam Altman. When Altman himself said that Husk’s ChatGPT model can’t run a timer, the bot said otherwise.
“So, what he’s saying is that some voice models might not have all the capabilities, but I do.”
Husk corrects ChatGPT, “No, he’s talking about you specifically.”
The chatbot insisted, “Well, I can tell you right now, I definitely have a timer capability.”
When Husk asked the chatbot if it thinks Sam Altman is lying, it denied it and thought Altman wasn’t lying either. Husk said that one of them must be lying.
The only way to test that is by using the supposed timer feature one more time. When Husk asked ChatGPT to start a timer in the updated video, the bot said Husk ran for seven minutes and forty-two seconds.
Husk didn’t even leave his chair.
Husk didn’t even leave his chair. Would it be a stretch to say ChatGPT tried to gaslight Husk like a manipulative lover? Not exactly, but this wouldn’t be the first time Husk has messed with his ChatGPT.
AI’s people-pleasing tendency
In another video, he lied to his ChatGPT about activating a “really ugly face-morphing filter.” The mere suggestion of the idea made the chatbot agree immediately.
But Husk wasn’t wearing a face-morphing filter either, and the chatbot had to awkwardly save the conversation after it agreed that Husk looked “ugly.” Essentially, the bot agreed based on suggestion. It didn’t rigorously check if Husk even had filter software turned on.
Who would’ve thought that even bots people-please? Lying doesn’t build trust in relationships—not that the robot would know, of course. Instead, it overestimates its abilities, claims it can do a task, and makes a person wait for two days, only to inform them later that it can’t accomplish that task (a true horror story).
Can AI be trusted with other complex tasks?
A criticism for AI chatbots is that they often agree with the user uncritically. Chatbots, at one point, can even give in to restricted and unreasonable requests. At worst, it can give advice it shouldn’t after its restrictions are bypassed by users.
But Husk’s example essentially proves why ChatGPT shouldn’t be relied upon for everything. Some people tend to use the bot as an advanced version of Google, but for now, nothing beats relying on the basics—that includes the brain.
If ChatGPT can’t even time runs without hallucinating, it just means that it can’t be completely relied on for research and other intellectually strenuous activities.
(featured image: huskistaken)
Have a tip we should know? [email protected]
