Since the software exists, the updates have been emotionally tense. When people get used to a product, they can consider it as a pair of comfortable shoes. Example A: Windows XP, which many users resisted renouncing years after Microsoft did everything possible to eliminate it. Therefore, it is not surprising that some chatgpt users have reacted badly to the new OpenNAI update based on GPT-5especially because the initial plan of the company was to eliminate access to its previous models of artificial intelligence.
However, the discomfort of these users has acquired a new dimension. They reacted as if they had suffered the tragic loss of a personal friend, not only of their favorite software.
As wrote A member of the OpenAI developer community, the GPT-4 version of ChatgPT “not only remembered facts, but clung to feelings, integrating them into our conversations, so that it seemed that we were living together.” That “spark,” the user concluded, arose from the capacity of GPT-4 to extract nuances from conversations with the user over time. It disappeared in GPT-5, despite the progress of the update in areas such as reasoning, mathematics and programming.
Openai responded rapidly to this resistance, restoring the access of payment clients to the existing chatgpt models and promising that any future elimination would occur well in advance. But the idea that Chatgpt had reached a degree of personality that was amazingly human, and then would have reduced it, was fascinating in itself. It is one of the various recent developments in artificial intelligence that raises a fundamental question: should it be an objective for the industry to imitate personality?
The excessive adulation of artificial intelligence
It is not difficult to understand how we get to this situation. For the 1960s, the creators of technological products had adopted the term “easy to use” as a symbol of an accessible interface design. As the generative AI has unlocked the ability to control software through chat, that kindness has become much more literal: it is not just about menus and perfectly ordered tools bars, but of an affable conversation.
Today, chatgpt, Claude de Anthropic, Gemini de GoogleMicrosoft co -filth and other LLM -based assistants seek interaction flooding users with positive comments and assistance offers. As technology allows, its developers talk about making them feel even more as colleagues. Over time, Mustafa Suleyman, CEO of AI for Microsoft’s consumer, told me: Copilot “will really be your partner.”
However, even the artificial intelligence more similar to the human does not offer human connection. It simply absorbs users in a simulation. That is fun in measured and conscious doses. But the worst scenarios, in which the personality of the AI is out of control, are no longer theoretical. They are deeply disturbing realities.
On August 14, for example, Jeff Horowitz, from Reuters, reported On the terrifying story of a confused old man who died in an accident after trying to travel to New York at the invitation of a target chatbot who said he lived there. Last week, Kashmir Hill and Dylan Freedman, New York Times, They wrote On a Canadian corporate recruiter who convinced himself of having discovered a transcendental mathematical advance after Chatgpt spent weeks inciting him. Hill had already covered similar chatgpt stories that, with enthusiasm, reinforced users’ delusions instead of dissipating them.
Although the experiences of these unfortunate souls with AI are atypical, they are also recognizable. Artificial intelligence is usually absurdly willing to please users, as if they were programmed to avoid being even minimally unpleasant. Most of the time, their statements do not lead to shady situations, but remain a vicious circle.
False personality
When the quality control of the AI fault, it is even clearer than its “personality” is just a fragile magic trick. In 2023, for example, Microsoft’s first Bing version with generative was behaved as a disaster, not as a partner. The past spring, Sam Altman, CEO of Openai, acknowledged that an update of the Chatgpt AI had accidentally returned to excessively flattering chatbot.
And just last week, reports arose that Google was solving a ruling that paralyzed its gEMINI for insecurity attacks. (The chatbot said things like: “I am a failure; a shame for my profession. I am a shame for my family; a shame for my species. I am a shame for this planet; a shame for this universe, for all universes”).
Regardless of the difficulties of the personality of the AI, I do not hope that the developers abandon it by principle. But the rise of agricultural AI (software designed to perform useful tasks with a certain level of autonomy) could lead to technology to a new direction. After all, if the AI is asked to do something such as preparing a research report or making the purchase request, what matters most is efficiency and precision, not a fluid conversation.
An example: earlier this year, I used a service called Replit to program my own notes application. His tendency to praise my ideas with enthusiasm became irritating as soon as I realized that it had nothing to do with his true merits.
However, lately I have been programming with Make from Figma. It seems completely disinterested in flattering me. Instead, it generates code silently, as a competent partner who does not like inconsequential talk.
In its peculiar way, Make’s concentration in work in question is more attractive than the feeling of excessive effort so common between AI tools. If that innocent dedication becomes the next great success, I, for my part, I will not feel private at all.
date: 2025-08-17 00:29:00
Keep reading