If it's just a "toy" then how is it able to have all this economic impact?
FaceDeer
joined 2 years ago
A huge market means there's lots of demand for the products. That doesn't have to translate to lots of jobs for the people producing that product.
So Meta's AIs will mainly reflect non-EU cultural values.
Let's say that your computer has the IP address 1.2.3.4. When you register for a DNS name, let's say bolexforsoup.com, you tell the DNS registrar to associate that name with your IP address. So later when my computer wants to communicate with your computer it asks the DNS system "what's the IP address for bolexforsoup.com?" And it tells me "1.2.3.4", which I can then use for communicating. The DNS service is not something you're running yourself, it's a service that someone else is running. That's the problem here. Your computer can be completely 100% FLOSS, you can be a master programmer who can manipulate your computer at will, but if my computer wants to talk to bolexforsoup.com the only way it can know the IP address for it is to ask DNS for it. That happens outside of your control. As we're seeing in this case with anti-piracy laws, this is something that an outside force - a government, a company, maybe even a lone malicious hacker - can interfere with if they want to stop me from reaching your computer.
It isn't a problem when you're just running software on your own computer and have no need to communicate with anyone else.
But that's not the case for domain names. It wouldn't work at all if we each had our own private little parallel universe, it defeats the whole purpose of a domain name system. We all need to agree on which names are associated with which IP addresses.
I'm not trying to promote blockchains as a one-size-fits-all universal solution for every problem. That's silly, no technology is a universal solution for every problem. Blockchains are very good at solving a specific subset of problems, and DNS names IMO is one of those. When you need everyone to agree on a particular fact and you don't want to designate some particular authority to be "in charge" of validating that fact then that's exactly what a blockchain is for.
How does your FLOSS software solve the Byzantine Generals problem? If two different people want to use the same domain name, how is it determined who gets it? These are the things that blockchains contribute a solution to.
It's not enough that the software that everything's running on is free/libre. Determining who gets a scarce resource (unique names) is the real difficulty here.
Full decentralization and censorship resistance. In the case of DNS services there's still an organization of some kind that you're having to trust to not mismanage your registration. Both now in their current form and in any future form the organization may take.
ENS, on the other hand, is just a smart contract running on Ethereum. Its behaviour is programmed, not dependent on any human decision making. To censor it you'd need to block Ethereum as a whole.
I just did. The ENS system, a decentralized replacement for DNS. That's what started this subthread.
Even if you trained the AI yourself from scratch you still can't be confident you know what the AI is going to say under any given circumstance. LLMs have an inherent unpredictability to them. That's part of their purpose, they're not databases or search engines.
if I were to download a pre-trained model from what I thought was a reputable source, but was man-in-the middled and provided with a maliciously trained model
This is a risk for anything you download off the Internet, even source code could be MITMed to give you something with malicious stuff embedded in it. And no, I don't believe you'd read and comprehend every line of it before you compile and run it. You need to verify checksums
As I said above, the real security comes from the code that's running the LLM model. If someone wanted to "listen in" on what you say to the AI, they'd need to compromise that code to have it send your inputs to them. The model itself can't do that. If someone wanted to have the model delete data or mess with your machine, it would be the execution framework of the model that's doing that, not the model itself. And so forth.
You can probably come up with edge cases that are more difficult to secure, such as a troubleshooting AI whose literal purpose is messing with your system's settings and whatnot, but that's why I said "99% of the way there" in my original comment. There's always edge cases.
Evidently "precision" isn't needed for the things the AI is being used for here.