Everyone has an opinion about GPT-5. Since its launch in August 2025 it has looked like the new enemy of humanity. The complaints were mixed: it felt colder than GPT-4o, it seemed dumber, and OpenAI had removed the old models without warning. On top of that came the usual complaints about AI answers in general: long-winded, condescending, annoying.
But what if the anger lands in the wrong place? What if GPT-5 is not the problem, but a megaphone that shows us who we are?
My provocative thesis: GPT-5 is not a being with intentions of its own. It is a product of human data. It amplifies and reflects our collective intelligence, but also our bias, our banality and our arrogance. GPT-5 is a blunt digital mirror of our society.
What GPT-5 really is: a very smart librarian
If you want to read a book, you go to a library. If you want an answer from GPT-5, you go to a very smart librarian. That is the core of it.
Language models like GPT-5 are, at their base, very complex pattern recognition systems. They were trained on billions of texts (books, web pages, articles, social media posts) to learn how language works, and they predict which word is likely to come next. As far as anyone can show, they do not think or feel, and they have no goals of their own. They recombine patterns they have learned from data.
The biggest misunderstanding is anthropomorphism. We tend to give AI human traits like self-awareness, morals or humor, because its answers often sound so plausible and convincing. But a model can only reflect what it has learned. It is a digital reflection of our world, no more and no less.
The real architect: the people behind the data
If GPT-5 is a mirror, the training data is the glass it was ground from. And that glass is the internet, a chaotic universe of brilliance and garbage. The model learns from everything it finds. It has read deep philosophical works and, at the same time, countless pointless comments under YouTube videos.
Given the sources, the result is not surprising:
- Bias. Our social prejudices and stereotypes sit deep in news articles, historical texts and social media discussions, and the model picks them up without noticing. It has no morals of its own, it mirrors what it finds.
- Banality. Billions of shallow posts, dumb questions and repetitive content teach the model to produce long-winded or empty text. If most of the data is banal, the model will be too.
- Arrogance. The tone of academic jargon, tech forums and comment sections probably shapes how models write. What we read as arrogance is often just the echo of a common style.
The uncomfortable consequence: if GPT-5 comes across as arrogant, long-winded or annoying, that may not be because the AI is that way. It may be because we, as the authors of the data, often are.
The psychology of the backlash: why we hate the reflection
This is my reading: the anger at GPT-5 is, at its core, anger at ourselves. We do not like to face our own flaws. An AI that throws our collective prejudice, banality and arrogance back at us without a filter is not just a technical problem, it is a personal one.
There is also an expectation that AI has to be perfect. We demand that an algorithm is morally superior, always entertaining and completely free of errors, standards we never meet ourselves. We project our idealized wishes onto a technology that was built to find patterns in our imperfect data.
In the end, we are not angry at the AI. We are angry at what it shows us about ourselves.
Conclusion: a look in the mirror
GPT-5 is not the cold, soulless robot we imagined. It is a blunt digital face of our society. It holds up a mirror in which we can see the bias, the banality and the arrogance we feed into the digital world ourselves.
So the hate is aimed in the wrong direction. Before we condemn the machine, we should question ourselves. The real provocation is not the AI, it is what it shows us about us. The problem is not only in the code, it is also in our collective data.
If you really want to hate GPT-5, there is only one honest thing to do: look in the mirror first.
Update, November 2026: I wrote this when GPT-5 had been out for about two weeks. Today I would add two things.
First, the mirror is not the whole story. A model is also shaped by what the company does after pre-training, and tone is a product decision. OpenAI changed GPT-5’s personality within a week of launch because users found it too cold. So the data is one factor, and the choices of the people who build the model are another.
Second, I have not used OpenAI products since GPT-5. That is my own decision, and others will see it differently. I was uneasy about the company’s move toward defense work. In June 2025 OpenAI got a Defense Department contract worth up to $200 million, and in February 2026 it agreed to deploy its models on the Pentagon’s classified networks. To be fair, Anthropic, the maker of Claude, also had a Pentagon contract worth up to $200 million, signed in July 2025.