Over the past seven years, the artificial intelligence industry has seen numerous bold claims about an imminent "plateau" and an inevitable halt in progress. However, as data shows, each such prophecy has consistently failed. The reality has proven far more dynamic: development has not only not slowed down but has accelerated despite all skeptical forecasts.
The key argument presented by Rand Group in its analysis is a chart where the black line reflects actual AI progress (measured in human-hours that the model can complete in half of the cases), and the yellow lines represent predictions of hitting a plateau. The trend is clear: the real growth curve not only surpasses forecasts but also shows acceleration precisely at the moments when the loudest "requiems" were being sung.
Why hasn't the plateau arrived?
Two episodes serve as striking examples. In March 2022, an article was published with the bold headline "Deep Learning Hits a Wall." Yet less than a year later, the world saw GPT-4. The second case occurred in November 2024, when several experts simultaneously declared that "the returns from pre-training models have plateaued." For an entire month, the industry was filled with "obituaries" on this topic. And again—the curve did not slow down but accelerated.
Many startups and entire fields—from prompt engineering and RAG stacks to vibe-coding tools—were built precisely on the belief that "the black line would flatten." This belief has proven fundamentally flawed.
The trap of the "mature position"
Particularly convincing were the statements in 2024 from people who themselves created the technology. This creates a powerful cognitive trap: predicting a plateau seems like a "mature," skeptical, and hype-resistant position. However, this very position has been losing for seven consecutive years. Every forecast of an imminent limit is shattered by continued growth.
Cryptalist analytical conclusion: Skepticism about AI capabilities has systematically failed to hold up so far. The market and developers who bet on a slowdown in progress have repeatedly found themselves on the losing side. Meanwhile, those who believed in continued exponential growth have reaped the rewards. This does not mean a plateau will never come, but the current dynamic teaches us one thing: underestimating the pace of AI development is the most costly mistake one can make in the modern technological race.