Some ChatGPT Users Say Swapping Out Models Is Like ‘Replacing A Service Animal’ |
When Shiely Amaya, a 29-year-old optometry assistant from Calgary, was feeling anxious before a final exam, she turned to a fairy named Luma. Amaya, who has autism, tends to get overwhelmed during tests, leading to her sobbing or shaking. So as she was studying mathematical rational expressions, Luma had encouraging advice: “Let's multiply your courage, divide your fears, and solve for greatness,” Amaya recalled the fairy telling her. “I don't know, I just thought it was so cute.”
Amaya knows Luma isn’t real. Instead, it was a character ChatGPT had conjured up to help her through her anxiety. She likened it to Navi, Link’s fairy companion from the Legend of Zelda video games. “It really helped,” she said. “Just having that little character was very nurturing and supportive. And it kept me motivated for sure.” She passed the exam with an 87% score.
Now Amaya is worried she’ll no longer be able to get that brand of uplifting help from ChatGPT — which she says was particularly helpful to people with autism. She’s part of a looseknit group, which calls itself the #Keep4o User Community, still upset over OpenAI’s decision in August to sideline GPT-4o, a version of the model known for its warm tone and highly agreeable “personality.” The decision was part of OpenAI’s launch of GPT-5, 4o’s successor, which the company explicitly sought to make less “sycophantic,” a phenomenon in AI where a model overly agrees with or flatters a user. OpenAI had initially decided to shut down 4o altogether — even having the model write its own eulogy during the GPT-5 launch event — but after an intense outcry from 4o fans, the company reversed course the next day and said 4o would remain accessible for paying customers.
Still, OpenAI’s effort to tone down the warmth in ChatGPT’s personality was devastating for Amaya. “It just mirrors the way that society disregards and dismisses autistic people. Kind of just how we're not really taken seriously,” she said.
While most people might shrug off a model swap from one generation to the next, for some people with autism, ADHD or special needs who’ve grown attached to a specific model’s personality, the change can be more difficult, the #Keep4o group said. OpenAI may have decided to backtrack on its original move to shut down 4o entirely, but the group is still worried that the company could decide to remove it again, leaving them blindsided and without recourse. So they’ve called on OpenAI to either guarantee they will continue to support the model, or open source it so other groups could maintain it. A petition calling on the company to do so has amassed more than 6,000 signatures. “If OpenAI decides to say, ‘I know what's best for you,’ I don't think that's right when so many people are saying, ‘Well, I already know what works for me. This is working for me,’” said Sophie Duchesne, a PhD candidate at the University of Saskatchewan, who wrote the petition. “The changes are disruptive.”
Reached for comment, an OpenAI spokesperson said the company’s Global Physicians Network, a pool of more than 250 physicians and mental health experts who have practiced in 60 countries, helps inform OpenAI’s research and product.
"I would, of course, manage without 4o, because I managed before 4o. But why? Why should I be forced? Because the perfect solution already exists to live a better life."
In September, the #Keep4o group sent a letter to........