ChatGPT-5: A Classic Bluff Nobody Wanted to Admit

ChatGPT-5: Quantum Leap Promise, Patchwork Delivery

The arrival of ChatGPT-5 was sold as a paradigm shift. A year prior, OpenAI’s announcements hinted at an intelligence capable of human-like reasoning, integrating text, audio, image, and video, and ending traditional applications. Reality, however, proved different: a collection of small models (o4-mini, o3, etc.) that fail to justify the expected leap and have actually worsened the experience for many users.

What Was Promised vs. What Was Delivered

The original idea was for ChatGPT-5 to unify multimodal capabilities and advanced reasoning. But OpenAI released intermediate versions over the past year—GPT-4o-Mini, GPT-4-Turbo, GPT-4-Vision, o3, etc.—until version 5 arrived without substantial novelties. As one analysis notes, "those improvements are what they’ve been making during this time; if they had presented them all at once, the perception would be different." The result: ChatGPT Plus subscribers lost access to reliable models like o4-mini and o4-mini-high and encountered more restrictive usage limits.

Real Problems in Daily Use

Users with established workflows in previous versions reported that ChatGPT-5 not only failed to improve but ruined tasks that worked perfectly. For example, a chat specialized in enhancing literary dialogues, mimicking a specific writer's style, stopped working. The new model "didn't hit" even with the same prompts. Others noticed a drier, less creative tone. Safety integration—"another way of saying it doesn't offer politically incorrect information or reasoning," according to one comment—further reduced utility.

The Structural Problem: Transformer Architecture

Beyond unfulfilled promises, a technical debate underlies the issue. Several experts argue that real reasoning requires an architecture distinct from Transformers, and that simply scaling up models will lead to energy supply problems. "By making the model larger, the only thing they’ll achieve is causing us energy supply issues," the argument goes. Chain of thought and reinforcement learning are improvements, but insufficient for achieving general intelligence.

The Hype Business and the Bubble

This overpromising is compared to the dot-com bubble of 2000: back then, people said the Internet had peaked, and now the same is said about AI. Snake-oil salespeople—those selling miracles—generate unrealistic expectations, and when these aren’t met, people feel cheated. The tool is useful for specific tasks (summarizing, translating, searching), but it isn’t the miraculous software that can do everything. As one analysis states, "we sell miracles that don’t arrive, and the fault isn’t with the tool, it’s with the snake-oil salespeople."

So, What Should We Expect?

In the short term, AI will remain a useful but limited assistant, far from general intelligence. Grand promises of a paradigm shift may materialize, but not with this version. Meanwhile, media noise and excessive expectations will continue to generate disappointment. Perhaps the true bluff isn’t the technology, but the narrative surrounding it.