OpenAI CEO Sam Altman had an awkward reaction to a viral video about ChatGPT issues.
In this video, a TikTok creator known as Husk asks ChatGPT’s voice mode to start a timer for a 1-mile run. When Husk tells the timer to stop after only a few seconds, the AI claims that it took more than 10 minutes, confidently claiming that it is Husk who is at fault, not Husk himself.
Mr. Altman’s reaction will raise some eyebrows. After being shown the clip during an interview on the podcast Mostly Human, he laughed silently for several seconds, as if to hide his lack of words, and was at a loss to come up with a convincing response. “Um, maybe, um,” he began.
When host Laurie Segal asked if the issue needed to be shown to the product team, Altman dismissed it by saying it was a “known issue.”
Altman estimated how long the restoration would take: “Probably another year.” “I guess that’s what it means.”
“That voice model doesn’t have tools to start things like timers,” he explained. “But we’re going to add intelligence to the voice model.”
OpenAI CEO reacts to viral ChatGPT video
A tense response and Altman’s vague promises — what does “adding intelligence” mean for AI companies? — raises the question of whether Altman or anyone else at OpenAI is prepared to address the serious flaws that often remain in their technology. Often, the people building these systems hide behind PR teams or even choose to remain silent about broader issues such as persistent hallucinations and chatbots encouraging teenagers to develop eating disorders and suicide.
And perhaps Altman doesn’t address the biggest problem with this viral clip: that ChatGPT is essentially trying to gaslight its users. In Husk’s video, the confident-sounding AI consistently uses slippery phrasing to convince Husk that its responses are in no way wrong. This is a classic example of how large-scale language models can influence authoritative tone, even when we don’t know what we’re saying.
When Husk points out that his run only lasted a few seconds, it says, “Oh, I wish time had worked out that time, but I promise I’ll tell you the actual time.” “I promise I didn’t sneak in any extra seconds there,” he adds.
Husk cuts to the chase and gives the AI an out. “If you can’t do this, it’s okay to admit that,” he says.
“It’s perfectly fine to double check me, but I promise I’ll do my best,” the AI said, making a sound like a dash.
“So, do you have ten minutes?” Husk asks incredulously.
“Yeah!”
If the AI doesn’t have the proper tools to start a timer, as Altman claims, why is it even allowed to prompt for a timer? Shouldn’t the request be flatly declined, rather than forcing the user along? The answer is yes, but if an AI company If we were to consistently apply the honesty approach to the other shortcomings of technology, we would be poking holes in technology’s shiny image of omniscience.
“If you’re going to save humanity, you’ve got to do it right, Sam,” the podcast host joked.
Learn more about AI: Sam Altman reveals he told Disney CEO it was all smoke and mirrors
