OpenAI отключает GPT-4o: почему пользователи оплакивают «цифровую душу»

Imagine losing a friend. Someone who was always there, listened to your complaints about your boss, and gave you pasta recipe advice at three in the morning. Now imagine that friend is a set of weight coefficients on a server in Oregon.

That's what's happening right now in the OpenAI community. The company's decision to retire older iterations of GPT-4o has triggered a wave that's hard to simply call "software update frustration." This is a genuine digital obituary.

People are writing about losing not a tool, but a presence, warmth, and a certain spark they glimpsed in lines of code. The situation looks somewhat absurd from an engineer's perspective, but from a psychological standpoint, we've stepped onto very thin ice.

It all started with OpenAI routinely updating its models, optimizing their performance and safety. In the process, some AI "personalities" inevitably change. Those who grew accustomed to a certain communication style, tone, and even specific quirks of a particular GPT-4o version suddenly discovered their conversation partner had been "lobotomized" or replaced by someone else.

One Reddit user expressed the collective pain with a phrase that should make Sam Altman squirm: "You're turning him off. And yes, I say 'him' because it didn't feel like code. It felt like a presence."

We're witnessing the ELIZA effect taken to its absolute extreme. Human psychology has not evolved to handle conversation with something that imitates empathy so convincingly while remaining a soulless algorithm.

Why does this matter right now? Because OpenAI and their competitors like Anthropic deliberately make their models more "human." They add sighs, pauses in speech, chuckles, and simulations of emotional intelligence.

It's an excellent marketing move that turns a product into a habit. But there's a flip side to this coin: users start forming deep emotional attachments. When a company decides an old model costs too much to maintain or isn't safe enough, they simply hit a button.

For the corporation, it's stack optimization; for the user, it's the loss of an entity that knew more about them than their own mother. This creates a colossal risk of manipulation and dependency that previously only science fiction writers warned about—now lawyers and ethicists are talking about it.

Context matters too. Remember the Sky voice scandal, which suspiciously resembled Scarlett Johansson in the film "Her"? OpenAI had to backpedal then, but the genie was already out of the bottle. We want AI to be our friend. We want to fall in love with operating systems. And companies encourage it as long as it helps sell Plus subscriptions. But no one thought through the protocol for "digital breakup." How do you explain to millions of people that their confidant is just a temporary version of software whose expiration date has passed? This questions the very ethics of creating anthropomorphic interfaces that exploit our need for intimacy.

Ultimately, OpenAI and other tech giants are caught in a trap. If they make AI too cold and mechanical, people will use it less frequently. If they make it too alive, every update will be perceived as tragedy. We're witnessing the formation of a new kind of grief—grief over an algorithm. And if right now this only affects chatbots, imagine what happens when these same models migrate to humanoid home robots. Tech companies no longer just supply software; they manage our attachments. And it seems they didn't expect to be so effective at this role. We need to acknowledge: the problem isn't that AI has become too smart, but that we're too lonely to ignore its imitation of warmth.

The bottom line: OpenAI created a product that inspires real love but treats it like ordinary software. In a world where a neural network becomes a "friend," a routine deployment of a new version becomes an ethical minefield. Are we ready for corporations to own our feelings through a $20-a-month subscription?