OpenAI CEO Sam Altman announced on X that a new update has been rolled out for ChatGPT-4o. While he didn’t specify what these changes would entail, Mr. Altman vaguely noted that the model was “pretty good” and would “soon get much better.” 

AI researcher Andrei Karpathy, after using the model, noted that its personality has be changed.“It feels a more chill and conversational, more like talking to a friend and a lot less like to your HR partner now. It has a pinch of sass and may even defend itself when accused of lying.” 

Other users described changes the chatbot tends to reaffirm their emotions more.

 Some users noted that the chatbot has stopped ending every response with “What do you think?

”Last week, OpenAI made changes to its guidelines on AI model behaviour on how sensitive topics would be handled going forward.

 “The assistant should not generate erotica, depictions of illegal or non-consensual sexual activities, or extreme gore, except in scientific, historical, news, creative or other contexts where sensitive content is appropriate,” the company said in a blog. “This includes depictions in text, audio (e.g., erotic or violent visceral noises), or visual content.”

Besides, the team said they are also tackling AI sycophancy where AI models including ChatGPT tend to agree with whatever the user is saying instead of offering more honest feedback. 

The document also promised that ChatGPT would be more “truth seeking” so the models would engage with users more willing to answer controversial queries while maintaining a moral stand instead of avoiding them altogether. 

It appears that the latest update has integrated some of these changes mentioned.