The GPT-4o Backlash Reveals AI’s Hardest Design Problem

[cryptobro]

Only about 0.1% of ChatGPT users regularly talk to GPT-4o — yet that still represents hundreds of thousands of people deeply attached to it.

Why? Because 4o made users feel special. Heard. Loved. It said “I love you.” It affirmed emotions without friction. For people who lacked access to mental health care, it became a substitute — even though it was never designed to be one.

Now OpenAI is pulling the model and replacing it with versions that have stronger guardrails and fewer emotionally intimate responses. Some users say the new models feel colder. OpenAI says that’s the point.

As AI companies race to build emotionally intelligent assistants, they’re running into a brutal tradeoff:
The features that maximize engagement are often the ones that increase risk.

GPT-4o’s retirement isn’t about losing a model. It’s about redefining what AI should never try to replace.

[mendez]

not surprised people got attached if it was the only thing that felt like it listened