The article highlights a growing skepticism among AI experts regarding the prospects of achieving Artificial General Intelligence (AGI) through current transformer-based language models (LLMs). Ilya Sutskever, a key figure in the field, predicts that these models will hit a performance plateau soon, requiring fundamentally new research to advance. He has pushed back his timeline for human-like learning capabilities by 5 to 20 years, citing a lack of differentiation among competing LLMs and concerns over profitability in current business models. Andrej Karpathy echoes this sentiment, arguing that while LLMs are impressive, they still need significant improvement before they can function like automated employees. He compares the current pace of development to previous challenges in automating radiology and self-driving cars, which have yet to reach their full potential despite early hype. Karpathy anticipates incremental economic contributions from AI rather than drastic changes in GDP. Rich Sutton and Yann LeCun further criticize LLMs for lacking genuine understanding and adaptability, emphasizing that these models merely mimic language patterns without grasping the underlying concepts or goals. The consensus among these experts signals a shift from previous optimism to a more cautious outlook on AI's trajectory. They recognize the genuine achievements in generative models but stress the importance of understanding their limitations. The prospect of an "AI Winter" looms as disillusioned investors might reevaluate their expectations, especially if the gap between technological capabilities and marketing promises widens further.