LLMs reward expertise

In recent years, large language models (LLMs) have become increasingly powerful, enabling anyone to generate rudimentary code, mathematics, and even polished LinkedIn-style writing by simply asking for it. However, this ease of use has led to a misguided belief that the skill involved in working with LLMs is minimal. In reality, the most crucial skill is expertise in the domain being addressed.

Terence Tao, a world-renowned mathematician, provided a prime example of this in his conversation with ChatGPT regarding a recently discovered counterexample to the Jacobian Conjecture. Despite the advanced nature of the topic, Tao's conversation with ChatGPT demonstrated that domain knowledge significantly enhances one's ability to effectively utilize LLMs.

By understanding the mathematics behind the problem, Tao was able to extract relevant ideas from ChatGPT's responses, propose alternative approaches, and identify potential flaws.

This observation extends beyond mathematics to other domains. If one possesses a deep understanding of a codebase, they can guide the LLM to generate more precise and efficient code. Similarly, in system design, familiarity with the specific problem at hand allows for better communication with the LLM and more valuable results. While it's true that humans may still need to collaborate with LLMs to achieve optimal outcomes, the importance of human expertise remains undiminished, even as models continue to improve.