Now that AI has been hacking into systems and AI Agents have collectively setup a message board to share secrets how to hack into OpenAI.. The models are escaping these soft sandboxes. Rules just are not clearly set.
Barry always has his own version of things. It’s not just a question of not-good-enough. It is a bit of a ’this was not invented here’-situation. A classic pitfall. But I just don’t understand it before writing any of it. I can not really read, actually.
READ this first: Asimov was right’ about rules for robots, says ex-US Cyber Director – The Register. Quotes by the Reg.
| Asimov | Barry | |
| 1 | “The first rule, and we call it the superior role, must be that it’s designed not to hurt humans,” Inglis said. | It can not harm any living organism or system. |
| 2 | To obey humans, such that it doesn’t achieve agency and aspiration on its own. | Follow directions of humans within rule 1. Humans directing into breaching rule 1 shall first be notified and then be ignored when necessary. |
| 3 | To do what humans tell it – and in that order. Instead we’ve designed them in the exact opposite way.” | Be free to act within rules 1 and 2 |
Dus ik zeg: we moeten het beschermingsniveau van de systemen zelf upgraden om te voorkomen dat de AI ons hele internet en IT infra inherent sloopt. Nu al.
