<aside>
👋Thanks for dropping by!
If you have any questions or suggestions for this guide, please reach out to me via email.
</aside>
More recent large language models (LLMs) have evolved to a point of being intellectually useful. This guide is intended to help you make the most of these technologies without running the risk of academic dishonesty. This starts with understanding what these technologies are good at: LLMs are uniquely suited to help one see connections between ideas, notice blind spots or contradictions in their thinking, and prompt the person to think more critically about an idea. Intentionally using them will help you think more clearly and carefully about difficult material.
At present, LLMs can offer robust, seemingly thoughtful arguments on a range of topics, but the quality of their reasoning reaches a ceiling rather quickly. When pressed, the models frequently turn in on themselves: wildly misrepresenting things, arriving at absurd conclusions to arguments, or otherwise “hallucinating.”
When using AI models—and I want to strongly encourage you to do so!—you understand the technology as a spotter for your thinking. (A spotter, as you likely know, is someone you entrust to help you when doing heavy weightlifting.) Let’s think about this metaphor a little bit:
Below, I detail a few possible applications that an emerging writer might find themselves in, and how they might use AI constructively and with integrity to improve their thinking. These approaches encourage you to understand these AI technologies instrumentally: as tools for thinking that will push you to quickly find and work in the areas where we will most need humans in the future: solving interesting problems and leading. (Credit here to Seth Godin for this framing.)
In the early stages of thinking about a topic, most people will look to Google’s top results, or start with a few tabs of Wikipedia open—after a little while, if you’re lucky, you’ll have a sense of what you want to inquire into. LLMs have a pretty clear understanding of topics at the “Wikipedia” level, and you can enlist them in helping you probe your curiosity in useful ways. For example:
<aside> 🤖 I am trying to think about an interesting subject to research for an inquiry project on American political culture. Ask me a few questions to get a sense of what I am interested in, giving examples related of areas I might be familiar with. After I give you a response, suggest a list of topics that I might research. When you generate the list of possible topics, each one should be phrased in the form of an open-ended question to direct my inquiry and frame the discussion.
Models can quickly break down an idea, and summarize it with clarity and nuance. However, the prompt that you use can generate very different results. It’s not a bad idea to re-prompt your model a few different times on a subject to hone in on the right level of rigor to match your understanding. (Remember: your spotter is here to help you find your limits—encourage the AI to add more complexity to a summary treatment to get deeper into the subject faster!)
Below, you can compare the responses of GPTo when prompted differently on the same subject:
<aside> 🤖 Please generate a summary on the subject of political polarization.
🆚
Summarize the scholarship around the subject of political polarization in the United States, with an emphasis on more recent thinking and writing on the subject. Note areas of possible interest, where something is still unresolved or unclear.