The price difference is that google steals your data. That’s it. OpenAI steals data, ask for money to use most of their models, and buy even more data from other companies stealing user data (like google and SO). Also indexing web pages is not even the “stealing” part of google, it’s just not comparable.
Yes, training AI on user data for free then selling the end product is a reasonable thing to be concerned about. It’d be different if the product was free or the data was sold to them with user consent.
SO has announced a subscription-based service trained on user data for free, and not only there’s not even opt-out, they’re mass-banning users for trying to “opt-out” manually. Tell me one thing here that’s not completely fucked up.
I haven’t tried it myself (tho I’m planning to do so soon), but check Onju voice, it tries to do something kinda similar.
I hope someone tries do pull that on an echo dot. Good hardware, shit software.
Edit: Update with related links.
The Onju Github i forgot before, tho it’s linked in pcbway. It has instructions to set it up along with home assistant and even a matrix bridge.
Onju voice satellite is a different project using the same custom pcb. This one looks better integrated with home assistant and has an actual wakeword system (unlike og onju, which doesn’t have one by design). This one feels more like “better private alexa for home assistant”.