OpenAI revealed late Friday that it has updated its latest model to sound “warmer and friendlier.”
The highly anticipated GPT-5 rollout was, in CEO Sam Altman’s words, “a little bumpier than we’d hoped,” with some users even saying they preferred the earlier GPT-4o.
To address those concerns, OpenAI says it has introduced subtle tweaks designed to make GPT-5 “more approachable.”
“You might notice small, genuine acknowledgments like ‘Good question’ or ‘Great start,’ without slipping into flattery,” the company explained in a social media post, adding that internal tests show no increase in sycophantic behavior compared with the original GPT-5 personality.
At a recent dinner with reporters, OpenAI leaders tried to steer conversation toward the company’s future beyond GPT-5, but, as journalist Max Zeff noted, the troubled launch remained unavoidable. VP Nick Turley admitted the model initially came across as “very to the point,” but confirmed that the new update should make it feel noticeably warmer.