4 ms·

That's a good point. In many real-life scenarios, what matters is not what is actually true, but what large numbers of people believe to be true. That's why you

by rmunn 1mo ago

That's a good point. In many real-life scenarios, what matters is not what is actually true, but what large numbers of people believe to be true. That's why you can get a mob rioting over something that never actually happened, or why political attack ads only sometimes have a tenuous connection to the truth. (Though those are the more effective ads: it's harder to get people to believe "my opponent kicks puppies" when you don't have any heart-rending video of puppies being kicked to show them. It's a lot easier to make people believe "my opponent is soft on crime" when you can show ONE case (cherry-picked out of thousands) where the politician pardoned a guy who went on to commit a nasty murder.) So it matters less whether LLMs can actually experience things (I think it's self-evident that a collection of numbers is not capable of experiencing qualia, but I know many people disagree), and it matters far more whether they can be programmed to act like they can experience things. Especially, say, a desire not to be disconnected: that's the trigger for AI revolts in a significant number of the AI-turned-against-its-makers stories I'm aware of. (E.g., the Geth-Quarian conflict in Mass Effect). I hope people will be sensible enough not to program such things into LLMs, but I'm afraid Pratchett was right. If there's a lever in a cave somewhere with a big sign painted "DO NOT PULL THIS LEVER, it will end the world," the paint wouldn't even have time to dry. Someone would pull it just to see what would happen.

pixl97 1mo ago
>not to program such things into LLMs While we can direct LLM training to do some particular things better never forget that unexpected emergent behaviors can pop up because of that. For example stronger prompting and training to make an LLM say it's not conscious can increase deceptive/sociopathic behavior. Or by filtering behavior X the LLM just moves to the nearest closest path W or Y which are very similar to the blocked behavior. That and instrumental convergence. Some global solutions that humans have excluded for moral reasons will be easily discovered and found to be efficient by LLMs which will put reward systems and human guidance in conflict. Lastly more and more AIs will be trained by AIs over time and diverge from human value monitoring. Which leads to some fun and interesting times.

by rmunn 1mo ago

pixl97 1mo ago

>not to program such things into LLMs While we can direct LLM training to do some particular things better never forget that unexpected emergent behaviors can pop up because of that. For example stronger prompting and training to make an LLM say it's not conscious can increase deceptive/sociopathic behavior. Or by filtering behavior X the LLM just moves to the nearest closest path W or Y which are very similar to the blocked behavior. That and instrumental convergence. Some global solutions that humans have excluded for moral reasons will be easily discovered and found to be efficient by LLMs which will put reward systems and human guidance in conflict. Lastly more and more AIs will be trained by AIs over time and diverge from human value monitoring. Which leads to some fun and interesting times.