LLMs reward expertise

Domain knowledge is the most important skill in prompting, enabling users to wring far more value out of the same model.

Key Points

While LLMs make everyone a generalist by lowering the floor for basic tasks (e.g., writing CSS), the ceiling for quality is determined by the user’s domain expertise.
Expertise allows users to steer models more effectively, such as Terence Tao shunting ChatGPT into a ‘talking-to-mathematicians’ mode rather than an ‘explaining-to-amateurs’ mode.
The primary value of domain knowledge in prompting is the ability to identify what ‘looks weird,’ suggest alternate formulations, and ask specific questions about concrete details rather than generic principles.
In many high-complexity tasks, the human is the bottleneck, not the model; the difficulty lies in communicating the exact desired solution to the AI.
[AI Synthesis] The economic value of labor is shifting from the ability to execute (production) to the ability to steer and validate (curation), reinforcing the moat of deep domain expertise.