Sam Altman addresses GPT-5 backlash, assures warmer tone, fixes, and more user control

In a series of posts, he admitted the company had “underestimated how much some of the things that people like in GPT-4o matter to them, even if GPT-5 performs better in most ways.”

The remarks follow mounting complaints from users who say GPT-5 feels different in personality, with many claiming it lacks the “warmth” and “emotional intelligence” of earlier models.

Altman noted that users have “very different opinions on the relative strength of GPT-4o vs GPT-5” and acknowledged that a one-size-fits-all approach does not work for AI bots.

In the immediate term, OpenAI will focus on completing the GPT-5 rollout and making it “warmer.”

Outlining a longer-term plan for more tailored user experiences, Altman said, “We really need good ways for different users to customise things… I am confident we can offer way more customisation than we do now while still encouraging healthy use.”

With a touch of humour, he observed that some users love emojis, while others “never want to see one.”

Altman also cautioned of imminent operational hurdles. “We are looking at a severe capacity challenge for next week,” he said, adding that the team is working to optimise systems and expand capacity. He pledged transparency about “tradeoffs” the company may have to make, recognising that “not everyone will be happy” with the outcomes. “Thanks for your patience with us; we will continue to react and improve quickly!” he said.

Several immediate changes to GPT-5’s rollout were announced, especially for paying subscribers:

– ChatGPT Plus users will see their rate limits doubled, tackling one of the main complaints over usage restrictions.
– Plus subscribers will retain access to GPT-4o, with the duration depending on demand.
– A malfunctioning “autoswitcher” that temporarily made GPT-5 appear “way dumber” is being repaired to ensure the correct model answers more often.
– The interface will be updated to make it clearer which model is responding and to simplify manual activation of the new “thinking” model.