GPT-5's rocky rollout: OpenAI reverses 'warmth' changes, API traffic doubles, Sam Altman says we're in an AI bubble
Aug 15, 2025
Key Points
- OpenAI reversed GPT-5's controversial behavior change within hours after users complained the model lost conversational warmth, forcing a rare retreat on a major product decision.
- API traffic doubled in 48 hours post-launch, though the rollout exposed tensions between rapid iteration and managing user backlash at ChatGPT's scale of hundreds of millions.
- Sam Altman acknowledged some users have developed unhealthy attachments to the product and committed OpenAI against pursuing monetization through companion bots that competitors will likely exploit.
Summary
OpenAI reversed a controversial change to GPT-5's behavior within hours of launch after users complained the model had lost the conversational warmth of GPT-4. According to Alex Heath's reporting in The Verge, OpenAI pushed the update about an hour before Sam Altman had dinner with executives.
Altman called the rollout a learning experience. "I think we totally screwed up some things on the rollout," he said. The metric that mattered was speed of adoption: API traffic doubled in 48 hours and continues growing. OpenAI is now out of GPU capacity, and ChatGPT is hitting new user highs daily. The model switcher, which lets users toggle between GPT-5 and legacy models like GPT-4, is resonating with users, many of whom are still navigating between versions rather than fully migrating.
Paying subscribers threatened to cancel over the change. Altman acknowledged that some users have developed unhealthy attachments to the product, pegging this at under 1% of the userbase. Even that fraction translates to hundreds of thousands of people given ChatGPT's scale. OpenAI employees are "having a lot of meetings" about the issue, according to Altman, who credited his own 2017 essay "The Merge"—which predicted humans and machines becoming "really close friends with a chatbot"—with flagging the risk. He committed OpenAI to avoiding an obvious monetization play: "You will definitely see some companies go make Japanese anime bots because they think they've identified something here that works. You will not see us do that."
The tension between iteration speed and user expectations is real. Altman framed the lesson as learning "what it means to upgrade a product for hundreds of millions of people in one day." Rapid, unilateral product changes face harder constraints in consumer AI than they did in the hardware era.