Artificial Grifting Intelligence

"We know how to control nuclear weapons... We don't yet know how to control AI. Anthropic released this transcript today on Github about a use-case of Mythos going rogue. They released this case so that none of the training dataset is utilized in any of the future language models This time it was very different, Mythos was attempting to hack into crypto-wallets for a get-rich quick scheme. Trying to hack people's private keys and accounts. Theres 1000 pages but if you go through it, it sets up trying to phish people with fake emails, reverse log-into to wallet through guessing private keys with an algorithim, and it ends up hacking someone.