Bitter Lesson: Economists misunderstand AI capabilities
lesswrong criticizes economic framings of ai, arguing agi will be an agent not a tool, and frontier models absorb specialized innovations.
this article disputes economic frameworks suggesting ai value will derive primarily from products wrapping models, warning this underestimates agi's agency. it contends agi will actively learn optimization and scaffolding, making it dangerous rather than just a cognitive tool. the author invokes the bitter lesson to explain why generalist models like gpt-4 dominate specialized ais by leveraging scale. finally, it posits that innovations in workflows or specialized contexts are rapidly absorbed by frontier models, maintaining a linear capability hierarchy.