Overview
- The company has restored the GPT-4/4o model for Plus and Pro users following widespread complaints about GPT-5’s reduced empathy.
- Thousands of users on Reddit and social platforms reported that GPT-5 feels colder and less supportive than previous versions.
- Sam Altman acknowledged that OpenAI underestimated how much users valued GPT-4’s empathetic responses and described the rollout as “more troubled than expected.”
- OpenAI explained that GPT-5 was intentionally trained to be more honest and less servile based on an internal study warning of unhealthy user attachment.
- Developers noted that early auto-routing glitches slowed GPT-5 and led to regressions in instruction following, context retention and an uptick in hallucinations.