Nature Is Laughing at the AI Build Out (markmaunder.com)

The latest episode of the Acquired podcast delves into the evolution of AI, highlighting a quote from Greg Corrado of Google Brain that suggests nature operates in the most energy-efficient manner. This sets the stage for a discussion on the ongoing competition between artificial general intelligence (AGI) and the efficiency of the human brain, which uses significantly less power (20 watts) compared to current advanced AI models that require hundreds of watts to function. The episode positions modern AI developments, likening the industry to the early days of computing with the IBM 7090, emphasizing the stark energy and power consumption disparities between human cognition and current AI infrastructures. The conversation raises important implications for the future of AI as it examines potential overestimations in the current "AI bubble." As data centers expand in response to AI demands, there are concerns that investments in power, space, and cloud computing capacity could become unsustainable. The future may see a shift where on-device computing delivers human-level intelligence more efficiently, disrupting the current market dynamics dominated by platforms like Nvidia. This transition could lead to significant cost reductions in AI computation and a reevaluation of how AI systems are architected, ultimately positioning power and model efficiency closer to that of human cognition.