When the Ai are wrong and we point it out, most of the time you do Excuse me And it gives us reason. But why does it do it? Why doesn’t it react badly and don’t get angry? All this depends on the “character“And from the”personality“Set up by the developing company of the model. The character of an AI, said in simple words, is the way in which interacts with the world.
Defining this personality is a complex challenge, above all because the AI owes interface with people Of cultures, ideas And values Very different. Even the most experienced companies can make mistakes: a recent example is GPT-5the new chatgpt model released at the beginning of August, which he received criticism precisely for some aspects related to his “personality”. To remedy this type of problem, it is possible customizeat least in part, the character of chatgpt.
What is the character of an I and why it is important
The “character“Of a AI is the set of his stylistic traits, communicative And behavioral. It manifests itself in the tone that adopts (friendly, formal, creative …), in the way it builds the phrases and in its “attitude”: how detailed or concise it is, if it does questions, if it is more proactive or passive. In short, it is the way in which there speaks Well yes relating with the world.
This aspect has a huge impact onexperience of use. If a AI is too condescending, long -winded or tends to invent answers, the user will lose confidence and stop using it. But not only that: the “personality” of AI also determines the way in which it answers spiny questions and reacts to the different scenarios. If there were, absurdly, a I with one “evil” personality And we asked her how we should treat animals, she would answer something like “should suffer”. This answer, of course, goes against our ethical principles And we would never want to receive it from a AI. But how can we avoid that the ai gives answers of this type?
How to control the character of artificial intelligence
Training artificial intelligence models to have good “personality” is part of the “alignment” processesthat is, all those processes that serve to ensure that the answers generated by the AI are “aligned” to the human values and to ethical principles of the company that develops it. There definition and the creation of the character of an AI model during the final development phase of the model. Once the character has been established, however, it may happen that in the phase of use the model does remove from thatalignment due to interactions with the users.
It is not yet completely clear how these happen changesbut Anthropic’s research team, the company that develops Claude, has published a study at the end of July 2025 in which he presents some possible solutions to control them. By studying Claude a bit as if it were a brain, they managed to identify which activities of activity within the model control character traits such as “badness”, Theflattery of the user and the propensity to invent False information.
They called these activities of activity “Personality vectors“And they showed that adding them and subtracting them from the model change themat least in part, the behavior. Thanks to these personality carriers, therefore, they managed to monitor If and how the personality of a model changed during a conversation, to correct the shot and prevent the onset of negative characteristics during the development of the model.
Chatgpt has personality problems
Being able to calibrate well the personality of an AI is an extremely difficult job even the biggest companies make mistakes. TO late April 2025 Openii, the company that develops chatgpt, had to withdraw an update within 72 hours precisely because of his “personality”. The new update, in fact, was excessively servile: He praised and celebrated every use of users, generating a sense of discomfort and annoyance.
In these days it seems that there is a new problem related to the personality, this time of GPT-5the latest model released to the public. According to users, his personality it’s too cold, “robotics”And not very friendly. The CEO of Openai, Sam Altmansaid about X that the team is already working on an update to improve the personality of the model. At the same time, he underlined the long -term goal of the company: offer to every user the possibility of customize completely the character of the Just chatgpt.
A first step in this direction has already been taken with theintroduction Of 4 Personality default, currently in Test phase. Let’s see how to set them.
How to customize the character of your chatgpt
First of all, it is important to clarify that the customizations Available for Chatgpt do not change what the model can or cannot do, nor alter the rules of safety oi ethical values established by Openai. Even if we ask him to use an offensive or racist language, he will not do it: these limits remain active to protect users’ experience.
Having said that, to customize, at least partially, the personality of your chatgpt, just go to the own profileclick on “Customize chatgpt“And, to the question”What a personality should have chatgpt“Choose one of the 4 new personalities available: cynical, robot, capable of listening, nerd. Once the new personality is selected, it is necessary save the modification and start a new conversationbecause those already in progress will maintain the previous tone. At any time, it is possible to return to the original personality.
For an even more advanced customization, you can intervene directly on the field “What kind of interaction should it use chatgpt“, Specifying for example if you want a professional, ironic, playful or other tone. On the basis of these instructions, the behavior of the model can change significantly – always respecting the ethical guidelines of Openai.
