Read it with your own eyes

Terence Tao's use of ChatGPT demonstrates that experts can prompt AI more effectively in their field, suggesting that domain expertise enhances AI utility. Herbert Lui argues that writers must read source material themselves rather than delegating context gathering to AI, as skipping reading leads to sloppier writing. Lui emphasizes that reading every line, as Robert Caro advises, enables better use of AI in writing.

When Terence Tao uses ChatGPT, he’s able to prompt it in ways people without his knowledge of mathematics aren’t able to. Here’s an example https://chatgpt.com/share/6a5fdc7a-d6f8-83e8-bbea-8deb42cfed56 . This suggests that if you’re an expert, you’ll be able to use AI better for that field than someone who’s not. Writing and editing is its own expertise. Part of that is being a good sponge of information. In order to fully tap into that expertise, you need to read the material. For example, if I’m about to write an outline for a technical blog post, and I have an interview transcript, a questionnaire, and a brain dump of information possibly written by AI , I find that I’m better able to understand what I’m writing when I read each document. That’s part of why I made my little reader app https://herbertlui.net/making-reading-docs-easier-on-the-eyes/ . I do eventually use AI, mostly to ask questions or find specific parts of the documents I might want to quote. But I don’t skip the reading. Robert Caro’s motto comes to mind, “ Turn every page https://www.nyhistory.org/exhibitions/turn-every-page-inside-robert-caro-archive .” Read every line. If you abdicate https://herbertlui.net/you-vs-the-parasite/ the context gathering to AI—like dumping all your source files into the LLM https://herbertlui.net/trust-summaries-ai-or-otherwise-at-your-own-risk/ , then asking it questions—your writing will turn out sloppier and shoddier. When you read the source material, you’ll be able to better write with AI or without it . August Lamm and Jimmy Daly have riffed on this lately.