What’s basically being said is, making an AI powered software local-only doesn’t make a difference and doesn’t matter. But that’s not true, and the arguments for that don’t seem coherent.
But the company hasn’t collected it, because it doesn’t have it. Your computer has it. So long as it stays on your computer, it cannot harm your privacy. That’s why there is such a big difference here; an actual massive loss of privacy that is guaranteed to be combined with everyone else’s data and used against you, vs a potential risk of loss of privacy from someone gaining unauthorized access to your computer.
Software that is designed not to send your data over the internet doesn’t collect your data. That’s what local-only means. If it does send your data over the internet, then it isn’t local-only. How is it still happening?
So you don’t think collection of user data is a meaningful privacy problem here? How does that work?
Even with Recall, a hypothetical non-local equivalent would be significantly worse. Whether Microsoft actually has your data or not obviously matters. Most conceivable software that uses local AI wouldn’t need any kind of profile building anyway, for instance that Firefox translation feature.
The thing that’s frustrating to me here is the lack of acknowledgement that the main privacy problem with AI services is sending all queries to some company’s server where they can do whatever they want with them.
I don’t see how the possibility it’s connected to some software system for profile building, is a reason to not care whether a language model is local only. The way things are worded here make it sound like this is just an intrinsic part of how LLMs work, but it just isn’t. The model still just does text prediction, any “memory” features are bolted on.
The use of local AI does not imply doing that, especially not the centralizing part. Even if some software does collect and store info locally (not inherent to the technology and anything with autosave already qualifies here), that is not close to as bad privacywise as filtering everything through a remote server, especially if there is some guarantee they won’t just randomly start exfiltrating it, like being open source.
I don’t care if your language model is “local-only” and runs on the user’s device. If it can build a profile of the user (regardless of accuracy) through their smartphone usage, that can and will be used against people.
I don’t know if I’m understanding this argument right, but the idea that integrating locally run AI is inherently privacy destroying in the same way as live service AI doesn’t make a lot of sense to me.
Nothing happened here except a small project website temporarily lagging. Nobody has been assaulted here, which makes this an extremely different situation than that.