Used as training data, or used as prompts to give further context? The former would be very troubling since it’d then be available to anyone able to engineer the right prompt. But I suspect they’re looking at doing the latter.
Tbh privacy aside the only way to get any LLM to sound like you is to train it on your data. If that would be possible locally I’d be all for it. Maybe not necessarily got SMS but certainly for work mail.
Used as training data, or used as prompts to give further context? The former would be very troubling since it’d then be available to anyone able to engineer the right prompt. But I suspect they’re looking at doing the latter.
Tbh privacy aside the only way to get any LLM to sound like you is to train it on your data. If that would be possible locally I’d be all for it. Maybe not necessarily got SMS but certainly for work mail.