I don't think AGI is right around the corner (2025) (www.dwarkesh.com)

A recent discussion highlighting the limitations of current large language models (LLMs) reveals a significant skepticism about the near-term emergence of artificial general intelligence (AGI). The speaker argues that while LLMs demonstrate impressive capabilities, they fundamentally lack "continual learning," which prevents them from adapting and improving over time like humans. This inability to process feedback and build contextual understanding hinders their practical application in white-collar environments, suggesting that even if AI development halts today, widespread automation may not occur as quickly as some expect. The conversation also touches upon the potential of future AI models that can learn from experience, predicting a transformative impact on the economy once continual learning is achieved. The speaker anticipates that, unlike today's models, these future AIs could learn and adapt across various tasks, rapidly approaching a form of superintelligence. However, they cautioned against expecting immediate breakthroughs, as significant research and technical challenges remain in integrating effective learning mechanisms into AI systems. This nuanced perspective underscores the need for patience and realism in the AI community regarding the pace of advancements toward more capable and autonomous AIs.