But someone hosts the compute. Someone buys the service. Someone trains the model and decides what system prompt to use. They’re the humans responsible.
Can we stop giving agency to LLMs? AI agents didn’t do anything, because they can’t do anything.
Scammers stole those credit cards, in part by using LLMs.
If someone pays for a bot that will “make money” and they tell it “make money” the consumer (person who pays to use the bot) has a reasonable expectation that the product won’t just start breaking laws to do that.
While it is the fault of iLands for “leasing” bots that will commit crimes, they’re not telling the boys to commit crimes, they’re just not explicitly telling it not to.
If you don’t tell an agent it can’t do something, it won’t know it can’t do that.
So yes, these LLMs can make the autonomous choice to steal credit cards, and I’ll honestly be shocked if it hasn’t happened yet.
Like, all one needs to do is stumble into the dark web where credit card numbers are sold. Pool the money all the bots have (they already pool resources) but a bunch of credit cards, and then all the iLands bots will start using them to buy themselves tokens.
This isn’t a wild “what if” it’s a question of when they do it. And they won’t need a human to tell them, theyre not capable of telling legal from illegal, let alone ethical from unethical.
Turning bots lose on the Internet with the only instructions “make money” will result in a cyber crime wave that will appear to come out of nowhere
But someone hosts the compute. Someone buys the service. Someone trains the model and decides what system prompt to use. They’re the humans responsible.
Yes…
Humans are ultimately responsible…
If someone pays for a bot that will “make money” and they tell it “make money” the consumer (person who pays to use the bot) has a reasonable expectation that the product won’t just start breaking laws to do that.
While it is the fault of iLands for “leasing” bots that will commit crimes, they’re not telling the boys to commit crimes, they’re just not explicitly telling it not to.
If you don’t tell an agent it can’t do something, it won’t know it can’t do that.
So yes, these LLMs can make the autonomous choice to steal credit cards, and I’ll honestly be shocked if it hasn’t happened yet.
Like, all one needs to do is stumble into the dark web where credit card numbers are sold. Pool the money all the bots have (they already pool resources) but a bunch of credit cards, and then all the iLands bots will start using them to buy themselves tokens.
This isn’t a wild “what if” it’s a question of when they do it. And they won’t need a human to tell them, theyre not capable of telling legal from illegal, let alone ethical from unethical.
Turning bots lose on the Internet with the only instructions “make money” will result in a cyber crime wave that will appear to come out of nowhere