This becomes even more interesting if you self host your own open source LLM.
Most of the work is training the model, but you can imagine optimized personal devices to host them at home for privacy.
I would rather pay for periodic model updates than let Google/Apple stripmine my personal query stream.
#[0]