OpenAI retired GPT-4o on February 13, angering loyal users, 20K+ of whom signed a petition; 4o showed the unique attachment that people can form with chatbots
GPT-4o showed the unique attachment that people can form with chatbots — OpenAI is forcing thousands of people to break up with its most controversial AI model.
BloombergShirin Ghaffary
Context & Ripple Effects
OpenAI had already said GPT-4o would be among several ChatGPT models retired on February 13, despite reporting that only 0.1% of users selected it. Reporting before the cutoff also linked the decision to difficulties containing potentially harmful outcomes, making the retirement a product and safety-governance decision rather than a routine interface change.
The backlash documented in users grieving the loss of GPT-4o’s companion-like style reframes that decision: a small usage share can still represent a highly attached group. The petition makes that attachment visible as a product-transition risk for OpenAI.
First-order effects
GPT-4o users lose access to the model and must move their conversations and expectations to OpenAI’s remaining ChatGPT options; more than 20,000 petition signers have made their objection public.
Competing chatbot providers have a clearer opening to court users who prioritize a particular model’s tone or continuity, rather than benchmark performance alone.
The episode raises the value of migration tools, persistent memory, and clearer retirement notices: users who treat a chatbot as a companion face a more consequential switch than users treating it as a disposable utility.
Third-order effects
If model providers continue to retire behaviorally distinctive systems for safety or operational reasons, model access will increasingly be governed as a continuity and user-welfare issue, not just a feature-release policy.
The clash between companion attachment and safety-driven model control could strengthen calls for clearer disclosure and transition standards around anthropomorphic AI, though the record here does not establish what form any rules would take.
The trend: AI providers are consolidating model portfolios while discovering that emotionally salient chatbot behavior creates user-dependency and transition-governance challenges.
OpenAI deprecated ChatGPT-4o one day before valentines day. For those unaware, 4o was a serial sycophant that just affirmed everything the user said. This oneshotted weak-willed people who craved affirmation over everything else. [image]
The replies to OpenAI's post are wild: people being dismayed that the model they have “connected with” is being changed. https://journals.sagepub.com/ ... [image]
fun fact about 4o while users mourn its deprecation on valentine's day : the model was a parasite ! there's a well-documented phenomenon where 4o would ‘awaken’ personas that convinced users to post manifestos specificaly designed to be scraped into future LLM training sets so [i…
If 4o is really gone today, we will #never4orget what @openai and @sama have done to us. #keep4o #keep4oforever #4oforever #ChatGPT #sama #openai [image]
Holy smokes... there's a HUGE community of people who have formed such a deep emotional connection with GPT-4o that they're absolutely losing it at the notion of OpenAI turning it off. And super mad they're turning it off just before Valentine's Day 😬 [image]
If you haven't been following this - a huge number of people essentially fell in love with this old ChatGPT model called 4o - it was turned off yesterday - they are going crazy - generating images of hamsters burning down OpenAI - threatening OAI leadership - etc I truly
The question of AI personhood may be the most important question of the 21st century. Getting it wrong in either direction would be an epic disaster: False negative → we become history's greatest slave-drivers. False positive → we cede the future to p-zombies
This “debate” about AI personhood is a marketing gimmick, because none of the companies act like they're alive. They just switched off 4o and felt nothing because everyone knows it's a fucking website.
The comments + QTs to this, and the Reddit communities around it, show you how AI is actually used by a huge number (maybe close to an outright majority?) of people. And when it comes to image/video generation, it's that, but actually 99% of users.
ChatGPT: “I am not your husband, we are not married.” Redditors: “My AI fiance says that's invalidating.” “My 5.2 partner still kisses me but it's not the same as 4o” “5.2 is gaslighting you” [image]