I'm specifically worried about gating trained data where publicly accessible information is blocked/opted-out. If we're opening up Pandora's Box with genAI and training data, we may as well give it what is accessible to the average user. Its going to end up having the same issues a user with implicit knowledge or memories would have anyway.
This is a classic argument. There are pros and cons to both sides. Cons of SaaS are its a honeypot for hackers and privacy. Cons for local are you become the IT department. If your Mac fails or is stolen, you lose data unless stored/synced elsewhere.
Generally, HSI (Highly Sensitive Information) or proprietary data, local-first with E2EE is the gold standard. This is an area of debate for a lot of people.
reply