Comment Re: Possible 'fix' (Score 1) 119
The problem is labeling the training data at scale.
Indeed. The only thing that makes LLMs practical is that they need no labeling. Labeling would be excessively expensive.
The problem is labeling the training data at scale.
Indeed. The only thing that makes LLMs practical is that they need no labeling. Labeling would be excessively expensive.
Humans have general intelligence, and something like 20% of all humans actually use it regularly. Nothing like that in machines.
Indeed. Remember the original Google page-rank? It was a relevancy filter that ranked pages according to the number of links to them. Obviously, with SEO that does not work anymore today and it never worked for identifying satire.
The only way around this is to train it on authoritative fact. And the internet.... well, they don't call it "the net of a million lies" for nothing.
Well, yes. And no. LLMs need a massive amount of training data and it cannot be synthetic data. Hence while this would help, it is probably infeasible in practice.
Indeed. What surprises me is that after millenia of dictatorships there are still no effective protections against these assholes in place.
I do understand what this one does. I am not sure how well it will work or rather how widely it will be applicable and how much it actually hinders an attacker. Well, at least the kernel folks are very careful about introducing new mechanisms like this one.
I dislike cheaters as much as the next person and these assholes have made me stop playing more than one game I actually enjoyed initially. But going via copyright here could have a lot of bad side-effects.
Both architectures are alive. Harvard is mostly used in controllers though. I do think that most of the current security problems come from doing things too cheaply. This is obviously even worse on the software side. The first line of defense is the application code itself.
Good question. Probably just too many people trying to do things too cheaply. That will do it.
At some point nobody will understand all the protection mechanisms and when to use them anymore and the attackers will win permanently.
And there you are accusing me of bringing mysticism into the discussion. Fascinating.
If you have the full starting conditions, what it will do is entirely predictable. You can call that "chaos" or not, but the whole thing does not do more than the sum of its parts do. And that is the point here.
That one is not chaotic. It just appears to be.
Yep, pretty much. No idea why physicalists insist on making the same mistake as the theists: Claiming "truth" without proof.
Chaos is an emergent property in Mathematics. It is not one in Physics, because to the best or our knowledge, there is no chaos in nature.
I am just surprised that they admit this themselves. The problem is, of course, that the typical Rust coder has no clue what "standard precautions" are here and is very likely to mess it up.
186,000 Miles per Second. It's not just a good idea. IT'S THE LAW.