Gary Marcus contends that generative AI is fundamentally unreliable because large language models are merely next-token predictors, not true reasoning systems, which is why they persistently hallucinate. He criticizes the AI industry for its narrow focus on this flawed technology, comparing it to climbing only one peak in a mountain range while ignoring other approaches. Marcus highlights the double-edged nature of generative AI, which can be magical for some tasks but also fabricates information and fosters delusions, leading to a growing public and governmental backlash. He argues that AI companies like OpenAI and Anthropic lack sustainable business models, burning billions monthly without profit, and that the myth of generative AI being close to AGI is a dangerous misconception. The discussion concludes with a call for more nuanced regulation and a diversified approach to AI research, contrasting the US's over-investment in LLMs with China's more balanced strategy.