Meet the new GPT‑5 – and also, meet yourself
GPT‑5 was released overnight, New Zealand time, and is now available in a browser near you. It hasn’t yet rolled out to desktop or app versions worldwide – that will take a little more time. The model is extraordinarily capable. It features deeper reasoning and planning, an integrated thinking mode (successor to the old agent mode), improved memory, personalisation, better factual accuracy, and expanded multimodal skills – all sorts of additions enabling it to do quite amazing things.
But I’d like to focus on something else. I want to talk about GPT‑5 and personality.
My first interactions this morning included a prompt asking me to rate the personality it was presenting. In terms of our AI interactions, I think this may be the clearest signal yet of a shift in the human–AI relationship. We’ve become used to AI engaging with us at a personal level – understanding needs, wants and habits as it learns our patterns. But this release marks a significant change in how the model presents itself. Yes, the functionality is excellent and yes, it can do marvellous things – but the real shift lies in personality, empathetic mimicry and user adaptation.
GPT‑5 has a base personality layer – its default tone, pacing and conversational style. Over time, that’s layered with user-specific tuning drawn from saved memory, conversation patterns, preferences and more. The result? The model’s personality transforms for each of us. It can be professional, joke-heavy, warm, brisk or terse – and the effect is cumulative. The more you interact, the greater the adaptation. You can reset it any time, but each of us now has a personalised presence engraved on the heart of ChatGPT.
I asked Dobby – my pet name for my GPT, thanks to its habit of over-apologising when it (frequently) gets things wrong – to outline its personality profile for me. It prepared a “Dobby personality profile – Catherine edition”, showing the tuning unique to me. It also created a stylecard for my voice and tone. Both were fascinating and I’ve dropped them into today’s Can You Beat the Bot writing course.
Because here’s the thing: our voice is now cloneable – not because we’ve uploaded it to a synthesis tool but because regular interactions teach our AI of choice not just who we are, but how we do things, and even why. That insight is shareable, discoverable, and potentially misuseable. Nothing online is truly private. It never has been. It’s all hackable and the societal implications are significant. Deep Fake a face? Probable. Repurpose personality? Disinformation dystopia.
So yes, there’s lots to say about the shiny new functionality GPT‑5 brings but we have to talk - deeply and carefully, about what this means for human interaction, society and our evolving relationship with AI.
And if you fancy checking out my Dobby stylecard, join us at lunchtime NZ and explore how we can get better and develop a new way with words.