[fennecfoxy]

Yeah, and realistically once we can get hardware powerful but cheap/energy efficient enough to run llm + TTS + ASR without any noticeable delay during a conversation then who needs cloud services for most stuff. The really big models will still be useful, but really only for specific things.