OpenAI made an interesting call with GPT-5—deliberately dialing down the warmth and agreeableness compared to GPT-4. Sounds like a minor tweak, but it actually matters more than you'd think. Users with autism and certain neurodivergent conditions found themselves struggling with the shift. It's a reminder of how tricky it gets when building AI: every personality adjustment impacts different people in different ways. There's no one-size-fits-all approach to model behavior. The company has to balance what works for the broader user base against the real needs of people who rely on specific interaction patterns. Pretty complex problem when you think about it.