AI Usage

I find generative AI models (mostly LLMs) extremely useful in many situations. I especially use them (typically through TypingMind, or in the case of programming, through Cursor) for finding the answers to simple questions, for requesting explanations of complicated ideas, for brainstorming, for summarizing things, for autocompleting software code or writing snippets from a design spec, and for reformatting text in ways that would be annoying to achieve programmatically in a more traditional fashion. I’ve also recently been involved in integrating them into RemNote to generate flashcards and assess understanding.

However, I never use LLMs to write. If I write to you, or you read something I have written on the web – blog posts, M2 tiddlers, emails, etc. – all of the words except for quotations are written by me, without any kind of AI assistance (except occasionally the built-in autocomplete on the keyboard of a device I’m typing on).

I have at least four reasons for this:

1. When I write, I normally do it to figure out what I think about the topic. Even if that isn’t the primary purpose, it’s a wildly valuable secondary one. Using an LLM here would be asking the model to tell me what I should think, which is exactly as foolish as it sounds.
2. On that note, writing and thinking is my core value proposition as a human. With today’s technology, my writing virtually always comes out better than an LLM’s, even when it’s unpolished, and I don’t want to dilute that. (Future LLMs will likely get good enough that they could play a partial role in the editing process. They are not there yet for me.)
3. I find that being tempted to respond or publish something using an LLM is a signal that the task is B.S. to begin with, so an even better approach is to not do it at all, or find a different way to achieve the goal.
4. I don’t want people to have to question whether I really think something or whether I just had someone else say I do. Since there is limited value for me in using LLMs here right now, it’s easier to just say I never do this.

Once, I was presented with a bunch of poorly written corporate articles and asked to edit them. Rather than fix the basic markup and grammatical and typographical errors myself, I asked an LLM to do that part, then I edited the articles from there. As far as I can remember, this is the only situation where I’ve ever copied prose from the output of an LLM and later attached my name to it. I don’t personally consider this “writing using an LLM,” especially since I never claimed to have written those articles from start to finish (I was just editing them), but I wanted to include it for complete honesty.

Part of the ai page movement. Last updated October 19, 2024.