Posted 2026–10–13.
I think it may be time to relax my policy of not using LLMs outside of work.
I’m not very happy about LLMs. I don’t think it’s any one factor, but a combination. Here’s a list:
- Many things about their training are unethical.
- Many of the effects of their proliferation and the absurd funding of the large “AI” companies are unpleasant.
- They have devalued (in the eyes of many people) various creative pursuits, many of which I love. While this does not stop me from appreciating them, I fear a negative effect on the already-ailing culture surrounding these pursuits.
- Related to that, they have made my employed work far less fun.
My sentiment toward LLMs is a kind of impotent, sad, mild rage-disgust. I have, up til now, abstained almost entirely from using these products in my personal life; I get enough of them at work.
However, I suspect this policy is of very little benefit to anyone, and is definitely not of any benefit to me. Regarding “anyone”, I mean I’m not saving the world by not using these products in some small capacity. Regarding “me”, I mean there are probably potential benefits to myself and my family if I did use these products judiciously. So, I suspect there’s a more nuanced version of this policy that is overall better for me, with a near-null negative effect on the general “rest of world”.
There are certain kinds of jobs where I think LLMs could benefit me:
- Tedious code refactors, where the level of complexity is beyond what I can do with mechanical tools.
- Complex or difficult searches.
- Jobs in the name of profit.
The first is the category of “jobs where I stand to learn nothing and have no fun”. If I won’t enjoy it or learn something, then I have no fear that I’m missing out on the benefits I’d get by doing it for realsies. In contrast, I will continue to refuse to use LLMs for more interesting programming, even where I may use them at work in the interest of producing more stakeholder value or whatever. I still have not actually done this in my personal programming projects, but maybe one day.
The second is in the category of “killer application of LLMs that I actually enjoy”. Google has been sanded down and pumped full of ads. Duckduckgo seems to lack content. Marginalia is intentionally limited in its scope. Kagi might be good; haven’t tried it. The LLM supports crafting an arbitrarily complex query with ongoing feedback (in the form of a chat) using what is a relatively small number of tokens. I cannot understand people who are using it for all their basic questions, but when I need to dig up something specific that I struggle to find unambiguous, unconflicting search terms for, it helps. This is basically the only application of LLMs I have used in my personal life.
The third is self-describing — it’s the category I continue to participate in, every weekday. I have repeatedly shown through my actions that I value getting paid more than I value my objections to LLMs. Therefore, it is silly to obstinately hold back from using them for personal for-profit projects, while using them (not necessarily happily) when getting paid by a company. Pick a lane, Jamie. I suspect if I was single, unindebted, and undepended-upon, I would have a different set of priorities here. I’m not sure in which direction.
After resolving this new policy and justifying (rationalising?) it to myself, I have a fear remaining: I fear that I will be tempted to expand my usage. LLMs appear to be associated with deskilling, and with progressive dependence. I have absolutely no desire for LLM usage to pervade my personal life. In fact, I’d rather not pay the companies a dollar for it — plausible for now, with my current level of usage, but will it last?