The author also questions the best way for a technical person to acquire domain expertise and whether it's necessary at all. They propose learning as you go or finding a co-founder in a non-computer domain as potential solutions. They invite readers to share their thoughts on these issues.
Key takeaways:
- Creating LLMs or AGI-style models requires massive compute and data, which startups are unlikely to have, giving incumbents a huge advantage when it comes to general AI.
- It's difficult for startups to create a moat with an API or similar, as incumbents keep innovating and creating their own services that often make these startups obsolete.
- An 'AI startup' would do best to develop domain expertise, create a useful product in that domain, collect data from users, and use the data to create a useful domain-specific narrow AI.
- Many software engineers want to create developer tools with AI, but this is the domain that is most likely to be oversaturated with AI tools, because AI people already tend to be developers who know about software development.