From prophet to product: How AI came back down to earth in 2025 (arstechnica.com)

In 2025, the AI landscape shifted dramatically as the initial fervor over large language models (LLMs) began to recede, leading to a focus on pragmatism within the AI community. While the past two years were marked by grandiose predictions of AGI and superintelligence, this year revealed a clearer understanding of AI's complexities and limitations. Companies now recognize that in order to monetize AI, they must create reliable, practical solutions rather than chase unattainable breakthroughs. For instance, OpenAI's CEO, Sam Altman, switched from asserting the company's roadmap to AGI to celebrating minor improvements in GPT-5.1's language capabilities, emphasizing the gap between aspirational goals and actual technical progress. The significance of this shift lies in the growing skepticism regarding the feasibility of earlier predictions, leading to a more sustainable and pragmatic approach to AI development. Notably, while the valuation of companies like Nvidia skyrockets, concerns about a potential AI bubble echo the early 2000s dot-com crash. Researchers and developers are now tasked with refining existing technologies and addressing fundamental flaws in current models, as the industry strives for functionality over fantasy in a rapidly evolving digital landscape.