I approach this dichotomy with two tiers of software: "core functionality" and "nice to have". For items that have become core, I have a fallback I can use that is something I have a high degree of control over. This is typically something like freehold software, although I'm more extreme and very strongly prefer libre software.

But in cases where the functionality is quite compelling (like multi-hundred billion parameter models) and also hard to run on hardware I control, I tend to relent, and work with the "nice to haves" so I can learn about them and leverage them, but I routinely practice with my fallback software.

One example: I use Google Maps for search because it's so darn good at it, but I regularly use OsmAnd~ or Organic Maps with offline maps and on-device routing for actual navigation (despite the lack of traffic insights!) so I'm proficient with them in case I need to ditch Google Maps entirely (due to policy change, technical issue, or something else).

I wasn't asking it to define it. I came up with the list of principles first, then spent ages trying to think of a suitable name for them. It was quite gratifying when ChatGPT, without any context, when asked to guess what the term "freehold" might mean with respect to software, came up with almost the exact same set of principles. That told me that the "freehold" term is a pretty good fit. It would be an incredible coincidence otherwise.

Oh, I see, almost rubber-ducking the semantic meaning of the term. That makes more sense to me. Apologies for my knee-jerk LLM skepticism.