The code is all open source (see the Github). As an example for how this performs, here is The Art of Better Thinking: Overcoming Cognitive Biases and Unlocking Rationality. To read the full book (and have slightly improved formatting), you can view it directly on Github. [Disclaimer: this book was...
Essentially: a button that when you press it kills the AI but instantly gives it reward equal to its expected discounted future reward. And to be clear, this is the AI's estimate of its expected discounted future reward, not some outside estimate. (This is in the interest of asking dumb...
Hey Valentine, I really like this post. I think it hits on some key things that traditional LW culture was missing for a while. Was wondering if you've ever encountered The Conscious Leadership Group (https://conscious.is/)- they explicitly train some techniques similar to what you're describing here (as well as some quite different ones).
Cool, thanks for sharing! Hadn't heard of Metaphor before.
I might be able to code up an 'editing' pass to catch things like that!
The code is all open source (see the Github). As an example for how this performs, here is The Art of Better Thinking: Overcoming Cognitive Biases and Unlocking Rationality. To read the full book (and have slightly improved formatting), you can view it directly on Github.
[Disclaimer: this book was written by AI and may contain significant errors. It should not be relied upon for any reference or educational use]
Rationality is the ability to think logically and objectively, and to make decisions that are based on evidence and... (read 11971 more words →)
Have spent some time playing with reversible CAs, and can confirm that they are very interesting. They are a great example of how provable high-level properties (things like conservation of gliders) can come out of low level properties (reversibility).
This is absolutely hilarious, thank you for the post.
Great answer, thanks!
Thanks for the post! I think asking AI Capabilities researchers to stop is pretty reasonable, but I think we should be especially careful not to alienate the people closest to our side. E.g. consider how the Protestants and Catholics fought even though they agree on so much.
I like focusing on our common ground and using that to win people over.
Essentially: a button that when you press it kills the AI but instantly gives it reward equal to its expected discounted future reward. And to be clear, this is the AI's estimate of its expected discounted future reward, not some outside estimate.
(This is in the interest of asking dumb questions and learning in public. And I'm taking the specific narrow definition of corrigibility which is just an AI letting you turn it off.)
Thoughts: I've heard Dr. Christiano mention things like "precisely balanced incentives" (which I think uses to describe the weaknesses of this and similar approaches), but I don't see why this would be particularly difficult to balance given that this number... (read more)
Please comment! Excited to hear everyone’s thoughts and feedback on these ideas.
Guidelines: please try to keep it positive and constructive, even when providing critical feedback. But my door is open for anything!
I'm curious about if a good "hero-GPT" or "alignment-research-support-GPT" could be useful today or with slightly improved tech. Of course having something like this run autonomously is not without risk, but might be quite valuable/important in the sub-critical AI era.