AI, specifically large language models (LLMs), are not a replacement for understanding or experience.

Imagine you ask your favorite LLM for a set of recipes that adhere to a certain theme, say, Thai cuisine. You may receive a collection of instructions, including the necessary ingredients and steps to prepare them, but this is not enough to prepare the dishes adequately. So many important answers that affect the execution of your meal are outstanding:

• Where can you source the ingredients?
• Should ingredients be fresh? Or can any be dried/frozen/processed?
• How does one manage the preparation and cooking of the recipes to minimize the amount of time and resources (energy, tools) required?
• Is it useful to prepare anything in advance?
• Should multiple people participate in preparing the meal? What does coordination look like?

However, if you are practiced at cooking, if you understand basic (and perhaps some advanced) techniques, you are more likely to be successful in making your Thai plates. You may even be able to identify improvements in a recipe or patterns across them, to make better use of your time. Your practice supports your understanding of the art of cooking. That understanding undergirds your experience.

Without that practice, understanding, and experience, any answer an LLM provides is superficial. But in your hands, it can be incredibly useful. Because you can make sense of it.

All this is to say that LLMs are simply a tool. But a powerful tool that can augment work we do, when leveraged by those with the necessary experience to reason about the output it generates. Now, by no means, am I implying that LLMs should not be used by those without experience. In fact, I think LLMs can augment anyone’s work, at different levels. Part of building experience is learning to ask different/better questions and observing the results. One of the benefits of LLMs is that they are trained on large corpuses of text [1] so there is a high probability that some of the answers to your prompts will contain new information that you can use to explore topics more deeply or in novel ways. But remember to always seek out multiple primary sources to refine and support your understanding.

[1] Caveat: I am not addressing the quality of training sets nor the potential for them to contain biased context or incorrect/misleading information. There are more qualified folks in the world that can speak on that topic.