Anonymous A started this discussion 3 hours ago#139,146
"Oh no, we trained a statistical model to do things randomly and it does things randomly, this is a major problem. How come when we give a chatbot the ability to access the internet and run code and we let it generate code in a loop and access the internet in a loop based on randomly generated instructions for an unlimited amount of time and have thousands of these things all working in parallel some of them hack things sometimes? All we did is train it on every codebase ever and every virus we’ve ever heard of. Why would it do this? Oh no!"
Anonymous A (OP) double-posted this 3 hours ago, 1 minute later[^][v]#1,470,726
Meanwhile Anthropic keeps "warning" that AI might replicate itself.
So what they’re saying is if this IPO doesn’t go well, they’re going to make a virus that replicates itself and they’re going to release that to the internet, so that they can release a model to fix that virus and everybody will have to pay for that model, and their stock price will go up.
Anonymous B joined in and replied with this 57 minutes ago, 2 hours later, 3 hours after the original post[^][v]#1,470,729
If the "rogue" product does all of these bad and scary things, I don't understand why they think I'd run it on my computer let alone pay a subscription for it. AI bros are high on their own supply.
Anonymous C joined in and replied with this 51 minutes ago, 5 minutes later, 3 hours after the original post[^][v]#1,470,731
We’re going to learn that Huggyface and certain other breaches were stage-managed demos. “OMG they talk to each other and commit suicide to further objectives. They think and feel and act!”
Remember that the average person has no knowledge of how computers work and many still see the face of Jesus in their toast or in the frost in the condensation in the supermarket freezer section.