The author proposes that we are at the end of the first wave of AI, characterized by scarcity and high costs, and predicts a second wave of AI by 2030, featuring new models, cheap GPUs, and open-source commoditization. To achieve this, the author argues for better tools and a shift in thinking about automation, suggesting that the most significant growth might occur outside traditional white-collar office work. The author concludes by emphasizing the need to move beyond "so-so automation" to achieve significant productivity improvements and address labor shortages.
Key takeaways:
- The U.S. workforce is facing a structural imbalance with too few people for all the jobs, largely due to factors such as demand growth, an aging society, retirements, lower immigration, and skill mismatches.
- Current AI technology is reaching its limits and is not enough to fill the gap in the workforce. The authors argue that we are at the tail end of the first wave of large language model-based AI.
- The authors propose that the next wave of AI, lasting until around 2030, will feature new models, ubiquitous/cheap GPUs, and a commoditization of LLMs. This new generation of AI will be lighter weight and more specific, potentially driving explosive productivity growth.
- However, to achieve this, we need better tools and a shift in our thinking about automation. The authors argue that we need to move away from so-called "so-so automation" that displaces workers without significant productivity gains, and towards "Zoso automation" that drives high productivity growth.