Before Large Language Models (LLMs), technical gaps required relying on colleagues or online resources. LLMs now enable generalists by delegating tasks like writing code or math. However, many mistakenly believe LLM interaction requires no skill, as anyone can ask for complex outputs. This overlooks the crucial role of domain expertise.

Mathematician Terence Tao's interaction with ChatGPT exemplifies this. His concise prompts and domain knowledge guided the LLM effectively, producing more focused outputs than typical user interactions. Tao didn't just follow the model; he pushed back, suggested alternatives, and identified inconsistencies, showcasing how expertise allows users to steer LLMs more precisely.

This principle extends beyond mathematics. In programming, familiarity with a codebase allows for more effective LLM guidance. By asking specific questions and leveraging existing knowledge, users can achieve better, more tailored results. While LLMs offer a baseline utility for those without expertise, domain knowledge unlocks significantly greater value, highlighting that human expertise remains a critical bottleneck in extracting desired solutions from these powerful models.