IMHO, this is how integrations with chat GPT should work. Default to a local model or a private cloud model, and if that doesn’t work, ask the user if they want the query to go to another LLM. And let people turn off the external LLM prompts entirely.
Let make it opt it. And make opting out very prominent.
Right - but try asking “set an alarm in five minutes” - this used to set an alarm (say it’s 10:00, would set a 10:05 alarm). Now it sets a timer, which of course snoozes differently