The author further argues that even if a startup develops the best version of a language model, it will struggle to maintain a lasting advantage due to the rapid pace of technological advancement in AI. He suggests that the real value lies in startups that operate in areas where data cannot be easily harvested from the internet, such as healthcare or scientific research. These startups can leverage AI improvements while owning proprietary, hard-to-get real-world data. However, he warns that most of the value generated by AI will either accrue to society or be captured by existing industry incumbents, with only a narrow slice accruing to new, young companies.
Key takeaways:
- Most AI startups, especially those formed post-ChatGPT hype, are doomed due to lack of differentiation and defensibility.
- The rapid advancement of AI technology makes it difficult for any single company to maintain a long-standing advantage in algorithmic prowess.
- The value of most AI startups is likely to be generated in areas where data cannot be easily harvested from the internet, such as healthcare or real-world experiments.
- While AI will generate a lot of value, much of it will accrue to society or existing industry incumbents, with a narrow slice accruing to new, young companies that can replace the incumbents.