First, there are people who meet the demand. Take the tech industry as an example—these people are posting at top conferences, leaving traces on social media, or have projects on GitHub/HF. Second, recruitment has a high likelihood of becoming outsourced because the difficulty of intellectual production has decreased. However, humans are indeed better at communication and alignment. Most people can’t clearly articulate their needs, which might lead to a bunch of super individuals taking on outsourced work in one place, with delivery cycles drastically shortened. Third, AI is indeed much better than humans at delivering single tasks. But when it comes to cyclical activities or inputs with variations, people still lack trust in the generalization ability of models. For non-standard tasks, the reasoning ability honed by large models in mathematical tasks is actually hard to fully adapt. This is precisely the significance of projects like @tydsh’s startup, which focuses on RL with non-verified rewards. This is the elephant in the room when it comes to human development and work demand.