A practical AI post... with recommended prompts.

I won’t belabor an introduction with an overview of how AI is received. Some love it, some hate it, some organizations ban it (though academia is coming around), and others highly encourage it (GenAI, anyone?). The fervor surrounding these issues is intense and will remain so for the near future. It’s so intense at the moment that I feel compelled to give a disclaimer like I did for my post on “Writing under the UCMJ.” I am an author with a couple of novels, so my disclaimer is this: I do not use AI in my creative writing. Why? Firstly, because AI is bad at writing, and will continue to get worse. Second, because I'm not writing to become successful. I just enjoy what it does for me. Giving that to AI would be like buying a self-driving racecar... a costly way to do nothing.

I will give a few practical tips for using AI. I’m keeping it simple and succinct, organizing by worst to best use cases based on my experience so far.

BLUF: AI is good at augmenting your thinking and bad at replacing it.

Bad Uses of AI: (Without the moral, ethical, or procedural arguments)

Generating writing. AI is wordy because common writing is wordy. If clear communication matters, if creative but deliberate application matters. Generating the actual body of writing will get you a solid “C” grade product at best. It feels lazy, pretentious, and uninspired. If that is okay for your project, then go for it. Where might it be okay? Formulaic memos, things that aren’t important, things that need information but will be locked away in a file for recordkeeping. But even if that is the case, never trust AI to do it right. Even generating boring, procedural memos requires a thorough review and correction before publication.

Generating outlines.

The more creative freedom you give an AI, the more room it has to hallucinate, fill the gaps, and generally go off script. If you start with nothing and ask it to create something, that is precisely what it does. In my experience generating outlines, you typically end up with a mid- to low-grade product, full of obvious points, clunky transitions, and leaps in logic.

Good uses of AI: (Without the fawning, cult vibes, and arrogance.)

Broadly critiquing your work. I like using AI to answer comprehensive questions about my

late-stage drafts. This is not to recommend changes, but just to see if you are coming across clearly and to check your tone. In this post, I discussed the three core questions you should ask yourself after writing a column: Do I have a point? What is it? Who cares? AI is a good azimuth check for these questions. Try asking things like

• “What is the author’s thesis?”

• "Based on this source, what are the author’s general feelings about [topic x]?”

• “Describe the author’s chain of logic.”

• “What are the strengths and weaknesses of the author’s argument?”

The final, critical step is to take the answers to all of these questions with a grain of salt. AI should augment and inspire your thinking, not replace it. If it says your thesis is something completely different from what you intended, take a hard look at it. But if it is close, trust yourself and your reader.

Reverse outlines: I mentioned above that outlining with AI is one of my least favorite applications. The opposite is true of the reverse outline. In final drafts, asking AI to summarize and outline your work is an excellent way to test the organization, clarity, and strength of your argument.

• “Summarize this document in an outline format.”

• “Outline the key points of this document using bulletized lists.”

Similar to the recommendation above, this is a “sanity check” only. Meant to identify potential problems to address.

An advanced search engine for research. Possibly my favorite use of AI is during the research phase. Tools that allow you to upload a library of sources and limit access to outside information (e.g., Google’s NotebookLM) save you hours. Instead of asking it to pull quotes or arguments, try asking:

• “Identify portions of these sources that discuss [topic x].”

• “Which of these sources talks the most about economics?”

• “Which of these sources discusses President Smith’s decision to invade Atropia? Include specific chapters and page numbers where possible.”

What I found was that I could target my research to specific sources based on what I was writing at the time. Key here, as always, is to remain only mildly interested in what the AI tells you about the source itself. Instead, go to the filtered list of sources and pull from there. You are not reducing the quality of your research; you are reducing the time spent finding it.

If you made it this far, I’ll summarize like this: Using AI to write is only slightly more helpful than asking a hammer to build a house. I’ve skipped the moral, philosophical, and speculative debates on purpose. (Though if you want a glimpse of my thoughts on free will and human-machine pairing, check out my Substack.) Use AI to inspire your own thinking, identify areas to explore in detail, and give you an “outside” view. AI is an excellent tool to augment your thinking. If it replaces your thinking, regardless of the aforementioned moral/ethical debates, you will end up with a worse product.