There's something to this argument. I for one use an API proxy (Kagi Assistant) when I'm submitting anything remotely sensitive to an LLM. But for coding in my hobby project, I have marked "use my conversations for training" in the Claude app. I do it in the off chance that one day we will use stupid amounts of compute for personalized medicine and to fight cancer; code for scientific and technical computing is really complicated and only a tiny fraction of humans can produce it.