A non-exhaustive list of principles I aim to follow with my use of AI:
Don't. With AI it's trivial to generate multiple (even hundreds) pages of content that support literally any idea. Doing so is roughly the equivalent of aSlop Grenadegish gallop, where you give so much (mediocre-to-bad-to-sometimes-okay) content that it's impossible for someone to effectively respond. For most things, shorter is better.Own what you produce and share. Don't make any arguments you don't understand. This is mostly for my research writing, but really for anything, you are accountable for what you share, even if you didn't write all the words. If AI cited a paper you've never seen, or a theory you've never read, it's time for you to do somegym work.Keep doing hard things. It's never been easier to start something. AI can build an 80% decent prototype with a single prompt. Polishing a project still takes significant time, which can feel even more frustrating because of how easy the first 80% went. It makes it all the more important to push past the easy 80% and spend time getting things production ready. Websites I normally would have either never built, or half built only for my use, now become products that I polish and publish. But I have to decide to put in the effort to do that. To sand down the rough edges and make something actually usable.Make things better. AI makes everything easier, including generating slop content to populate a blog to bump your SEO, or generating slop LinkedIn posts that "drive engagement." The things I build and write with AI are built with the aim of improving the world in some way. Making life better, and not just for me. If I don't see it as something that makes something better for someone, I'm better off not doing that thing.Don't fall for the sycophancy. AI's really good at telling you your ideas are amazing. Tell it your ideas and it will build it into a glorious bubble of praise, eventually turning it into a slop grenade if you ask for it (see rule 1). AI has no will, AI has no taste, AI has no opinions. It predicts the next token to create an answer it predicts will satisfy your request, and that's all. So when it tells you your idea is good, treat it as you would the opinion of somebody who always says your idea is good. (this applies even if you tell it to critique your ideas. Remember, its goal is to predict the next token, and that's all)- Corollary: It's really hard not to fall for the sycophancy.
- Corollary: Don't take credit for things AI did. I'm a little neurotic about this one, but I try really hard when I'm talking about ideas with other people to make clear what ideas were mine, and what came from AI. If I'm working on survey measures and use Claude to build some, I clearly acknowledge that when sharing the ideas with others. I think I do this for a couple reasons. First, because I want to acknowledge my own limitations. Second, because I want to help people feel more free to critique the output - if I say it came with AI's help, maybe they'll feel more okay about saying it sucks. I haven't found that to be the case, but that's the hope.You can be productive enough. Honestly, I think there's a limit to how much I need to get done most days. If I've been efficiently managing a bunch of agents to work on several projects at once for a while, maybe it's time to let them run for a bit while Igo to the gym, or just go hang with the family for a bit. It's easy to feel like if you're not using all your tokens that you're missing out or falling behind. But you're not. If you're using agents to accomplish anything right now, you're ahead of the curve and doing just fine. Most AI FOMO content is hype garbage and self promotion rather than actual productivity.