The provided text explores the current optimism and significant investment in AI reasoning models, which are believed to be the next step towards superintelligence. These models are designed to "think" by breaking down problems, similar to human cognition, and major tech companies are spending billions to develop them. However, new research, notably from Apple and other institutions, challenges this widespread belief, suggesting that what appears to be advanced reasoning may merely be pattern matching or memorisation at scale. These studies indicate that when faced with unfamiliar or complex problems, these models' performance collapses, revealing a fundamental limitation in their ability to generalise beyond trained data. This raises concerns about the long-term viability of current AI development strategies, the return on substantial investments, and whether the industry is overpromising the near-term arrival of true superintelligence.