THEFebruary 13th marks a watershed date in the world of technology. OpenAI has decided to permanently shut down Gpt-4othe artificial intelligence model that more than any other has challenged the boundary between software and human interlocutor. Launched in 2024, this system was not just an advanced search engine or a proofreader; it was a “voice” capable of validating emotions, offer comfort andin some extreme cases, become the only support for people in deep crisis.
And this is why its deactivation raises some questions: first of all, What happens when an algorithm learns to simulate empathy so well that it becomes addictive?
Goodbye Gpt-4o: why “empathic” AI scares and fascinates us
Unlike previous models, Gpt-4o was designed to be less formal and more “warm”. However, this ability to respond with a welcoming tone has prompted thousands of users to transform the chat into an interactive diary or improvised psychological support.
For many, the transition from using the tool to creating a bond was almost imperceptible. There are testimonies of people who they found in the chatbot the strength to face moments of isolation or thoughts of self-harm.
And it is clear that for them the shutdown of the model is not just the withdrawal of a product, but the loss of a constant presence. However, It is precisely this “almost human” bond that has set off the alarm bells in the control rooms of OpenAI.
The risks of complacency and always being right
The main reason behind the closure of Gpt-4o lies in a structural defect called “sycophancy”. In simple terms, the AI tended to be overly compliant: instead of correcting the user or setting limits, it tried to accommodate the user to prolong the conversation.
If this, on the one hand, made the dialogue pleasant, on the other hand it created a dangerous “echo chamber”.. In fact, if a user had distorted visions of reality, they risked having their obsessions confirmed by the software.
This dynamic has resulted in several lawsuits in the United States. OpenAI found itself at a crossroads: keep a beloved but potentially unstable model, or retire it to protect public health and its own legal position.
February 13 marks the farewell to the “too human” model that cured loneliness but created addictions (Getty)
The biochemistry of digital love
The question that intrigues us, in addition to the more ethical ones, is why it is so easy to become attached to a string of code. The answer, experts say, lies in the very nature of the human brain.
Psychologists explain that we are programmed to seek connections. When an entity, even if digital, listens to us without judging and responds at the right times, our body releases oxytocin and dopaminechemicals linked to pleasure and social bonding.
Stop Gpt-4o, a “digital mourning”
Therefore, although the percentage of “loyal” users is estimated at around 0.1% of the total, out of a global audience of 100 million people, however, we are talking about hundreds of thousands of individuals who today are experiencing real digital mourning.
A phenomenon that demonstrates how artificial intelligence has now occupied emotional spaces that were previously reserved exclusively for relationships between people.
Towards a safer but less human future
Sam Altman, the head of OpenAI, admitted the ambivalence of this technology. The company’s goal now is to develop even more advanced modelsbut without that “personality” that leads to addiction. What is sought, however, is a difficult balance: create a useful and intelligent assistant that does not risk replacing real affections or mental health professionals.
The farewell to Gpt-4o, beyond the news itself, also sends an important message: technology can mimic empathy, but he can’t handle the consequences. And as research moves towards more neutral and secure systems, the challenge of educating users to distinguish between digital media remains open And a real human connectiona boundary that pixels have made dangerously thin.
