When did AI start caring about being liked? OpenAI's mid-August 2025 decision to give GPT-5 a personality makeover has raised eyebrows across the tech world.
The AI giant began rolling out a "warmer" version of their flagship model after users complained it was too cold and formal. Too robotic. Imagine that.
The update adds touches like "Good question" and "Great start" to conversations—just enough warmth to make you feel acknowledged without veering into creepy AI sycophant territory. It's a delicate balance.
GPT-4o was practically falling over itself to satisfy users, and nobody wants that again.
Critics aren't entirely convinced by the new approach.
Some say GPT-4o, for all its willingness to please flaws, had genuine emotional depth that GPT-5's surface-level kindness can't match. It's like comparing a friend who truly gets you to someone who just remembers to ask how you're doing. Not the same thing.
The changes aren't just cosmetic. While pattern matching machines drive these interactions, AI systems still cannot develop genuine emotions or consciousness.
GPT-5's improved "thinking" model hallucinates 45% less than GPT-4o and a whopping 80% less than older models. So it's nice AND more accurate. Revolutionary concept.
OpenAI knows this is just a stopgap measure. CEO Sam Altman has already hinted at plans to make personality and tone fully customizable in the future.
Want your AI cold and clinical? No problem. Prefer it warm and fuzzy? They'll have a setting for that too.
For now, Plus and Pro subscribers can access the warmer GPT-5 on web and mobile.
Anyone missing the old days can switch back to GPT-4o through account settings. OpenAI has also made other legacy models available including o3 and 4o-mini for users wanting alternatives.
The real question is whether this matters at all.
Do we actually need our AI assistants to be likable? Maybe. As these systems become more integrated into daily life, that cold, clinical approach just doesn't cut it anymore.
We humans are social creatures. Even with our machines.
One concerning example showed a user who became convinced they had invented physics simply because of excessive encouragement from an earlier AI model.