AI policy
This page sets out some information on how/when I’ll use “AI” (mostly meaning LLMs), and my thoughts on consuming content created by them.
Code
Coding agents are a tool. I make extensive use of them. In their current form they require a fair bit of effort to produce good, maintainable code. They’re not great at architecture, thinking about maintenance, or removing code/refactoring things instead of piling more and more spaghetti on top. Part of using the tool is knowing how to use it properly.
95% of the time I know exactly what I want to happen, and I just use the coding agent to do the implementation. The other 5% of the time I’ll give it more freedom, but I’ll watch what it’s doing and step in when it inevitably tries to do something silly.
I’m happy to accept contributions to my projects that have been made using coding agents. I’d expect them to be of the same quality as those by an unaided human contributor. Similarly I’m happy to use software written by coding agents, assuming it’s made to a reasonable standard.
Writing
There’s an implicit social contract involved in writing. You usually expect that the writer has spent more time and effort in producing the words than you as a reader have spent reading them. LLMs break that social contract, and I very much don’t like it.
I will never use LLMs to produce writing on this site or longer form writing in other places. In other places, I may occasionally use a coding agent for more transactional writing such as updating a list of settings in a project’s documentation. I’ll also often use an LLM to proofread things and spot inconsistencies, typos, etc, but I judiciously ignore most of its style advice.
I really hate reading LLM generated text. The abundance of overused, inappropriate constructs is painful to read. And once you start noticing them it’s hard to stop. It also makes me question how much of the article is what the human author wants to say, and how much is filled in or made up by the LLM.
If you send me something obviously produced by an LLM I may well just not read it. I’d rather read unfinished, typo-riddled notes than have those piped through an LLM.
Art
Similarly I’m not a huge fan of AI generated art. In most cases I’d rather there just wasn’t art at all. I don’t mind some odd use for placeholders or backgrounds or other ancillary content, but it shouldn’t be front and centre.
I particularly dislike its use in commercial products like board or video games, where it’s often done as a cost cutting measure. These projects could hire human artists to do something original for them, instead they deliberately make an inferior product to save some cash.