How We Prevent the AI’s from Killing us with Paul Christiano

I somewhat misremembered; it looks like his point is less about mutual trust and more about supporting whoever has control of reward channels. It starts here where Christiano says that an AI takeover might follow the dynamics of a coup: ▶ this video And goes into more detail here: ▶ this video "Suppose that I've been tasked with helping defend you from some other AIs.... My job is, someone is coming to hack your computer and I'm supposed to help defend you. Supposed to help improve your security situation, whatever. And I'm wondering, what is it I could do that will get me a high reward. And one thing I could do that will get me a high reward is actually helping defend your computer, doing the task you actually asked me to do. But another way I can get a high reward is by saying at the end of the day what actually matters is just how you measure my performance. And your measurements of my performance ultimately are just entering some numbers into a dataset somewhere, something a computer says about how well I did. And it would really be much better if I were to just work with this AI who is attempting to attack you and say hey, AI who is invading, you know what, if you just help me, and we both make it look like I did a really good job, like I win, you win because you got the person's stuff; I'm going to get a really high rating because all the numbers that are going to be entered in the dataset are going to be really high, this is a win-win, everyone is happy." "In some sense what all

>Do you have a counterexample of someone who's gotten their predictions right? Because if not, that should only terrify you even more. If there's anyone out there who predicted how LLMs would work in a way Eliezer failed to, and if that person is predicting "AGI will be cool and will naturally prioritize our well-being", I would love to know. A recent interview with Paul Christiano is about the closest I've come to this. He does note some semi-accurate predictions at the linked timestamp, but the forecast for how things are likely to go is not exactly rosy, though he's quite a bit more optimistic than Eliezer. ▶ this video Also this whole interview was pretty interesting. Near the end he details how few people world-wide actually work on X-risk from AGI. He also outlines how the academic ML community in general just continually keeps getting predictions really wrong, and many aren't taking X-risk seriously. Overall his is the most balanced take I've seen. A lot better than Eliezer.

This quote is by Dr. Paul Christiano, PhD Inventor of OpenAI's Deep Reinforcement Learning from Human Preferences (RHLF). Start of thread: link Full video, "How to Solve AI Alignment with Paul Christiano": ▶ this video