Over the past few years, the world of artificial intelligence has been repeatedly buried. Every time another "expert" announces an imminent slowdown in progress, reality refutes these gloomy predictions. As analyst data shows, all such "obituaries" share the same fundamental error: a lack of understanding that development is not just continuing, but accelerating.
This is most clearly demonstrated by a graph where the black line reflects actual AI progress, and the yellow lines represent erroneous predictions of hitting a plateau. The key metric here is the task length in person-hours that the model can complete half the time. And this curve is steadily rising.
Loud claims that didn't come true
Several high-profile episodes serve as vivid examples. In March 2022, an article titled "Deep Learning Hits a Wall" was published. About a year later, the GPT-4 model emerged, which not only refuted this thesis but literally exploded the market.
The second episode dates back to November 2024. At that time, a wave of claims swept through the industry that "the returns from model pre-training have plateaued" and that the models themselves "have hit a ceiling." For an entire month, the industry indulged in mournful sentiments.
However, as the data shows, while these "farewell speeches" were being written, the progress curve did not slow down but, on the contrary, accelerated. Many projects relied on the belief in an imminent plateau: from startups in the RAG stack and prompt engineering to "wrappers" and web coding tools. All of them were built on the assumption that "the black line would flatten."
The trap of the "mature position"
Why were these predictions so easy to believe? The 2024 claim sounded particularly convincing because it came from people who were themselves creating the technology. This is precisely where the trap lies. Predicting a plateau always seems like a "mature" position—skeptical and immune to hype.
However, for seven years in a row now, this position has proven to be a losing one. Every prediction of an imminent limit is shattered by continued growth. The key conclusion is obvious: skepticism about AI capabilities systematically fails to be justified. Betting on a halt in progress repeatedly lets down those who make that bet.
Cryptalist's comment: This pattern is a crucial lesson not only for developers but also for investors. In a world where technology develops exponentially, "common sense" and conservative estimates often turn out to be the riskiest path. The market constantly overestimates long-term obstacles and underestimates the short-term potential for breakthroughs. Ignoring this pattern means systematically losing to those who bet on continued growth.