I saw somebody give a talk about this on the open source stage at the bitcoin conference this past week.
That sounds very intriguing but it also sounds like an even bigger risk than installing an AI on my hardware already.
What are the risks and rewards to this? What are the pros and cons?
I have a Mac mini sitting on my desktop that goes mostly unused unless I’m actively working. What sort of benefit could I get from my machine with open agents? What sort of risks would I be opening myself up to by using the service?
Same software surface risk you face when downloading pretty much anything that talks to the internet
If the device you would use it on has co-existing data you don't want to exposed to that surface you could boot into separate images, one for your stuff one for when it's idle... having your personal one encrypted.
That's not full isolation ofc, but about as good as you can do without a physical disk switch.
That's a good idea! I'm quite sure that all data is encrypted at rest by default on all M* Mac models so all you'd need is a second system partition. Never mount them at the same time and you'll have acceptable protection on anything that doesn't have chip vulns. So M3 and up is fine right now.
Yea the only obvious hole I can think of without thinking too much about it would be something infecting the boot loader, a physical switch would prevent that, but there's probably other ways to mitigate.
Yeah. Privilege escalation could turn off secure boot - at least all I needed was admin to turn it off last time I tested that. So that risk remains, but it is pretty obvious that this is happening when you boot, because it tells you and iirc it doesn't auto boot then.
I think if you have a spare device you aren't using it makes a lot of sense. I have not yet gotten comfortable with the idea of giving any AI control of computer.
I am kind of in the same boat, but I run my agent on dedicated hardware because I have the same paranoia.
I can’t run the risk of having it delete thousands of dollars of previous work.
I asked my agent what the risks are, and it said because you are technically opening up your own hardware to an AI, it could be used for something illegal and you are liable because it is your hardware doing that which is illegal.
Some of it sounded pretty intense.
pretty straightforward, have it running on my spare MacBook Air
but after working through it with the oa team, it became clear this is actually an important problem someone in Bitcoin should be trying to solve
and the best part is it’s open source
May I ask approximately how many sats one might expect to stack with such a setup?
that’s an m1, so far on the screen says 14k, still early days lots of bugs, but has incredible potential
yeah, it seems awesome but I have concerns about the liability and stuff like that. It’s not a problem with the company, per se. It’s a problem with law not being able to keep up with innovation and then coming after people like you and me because we ran some “illegal software“ on a spare computer.
There are definitely going to be some illegal battles surrounding AI just like they were cryptography.
AI will be considered “military grade weapons” and “need to be regulated.”
its entirely open source ideally the government could run their own version just like bitcoin, eventually this will be included in Autopilot, which is the next version
It's a security risk to have shared tenancy with anons on the internet on a device that also contains your work data. If there is an RCE in the server software and a privilege escalation bug in the OS, then you are screwed. Neither are uncommon phenomena and the threats are intensifying.