I know that the target audience of this article are very technical people, but it’s really dense and has a lot of technical language that is not common knowledge for some people who have expertise in things like alignment and evals, but not working in your narrow niche. (like me)
Also, I would love to see more information on how your findings connect with the broader things that AI safety people should care about.
I know that making research is hard, and hope that my comments are helpful.
I know that the target audience of this article are very technical people, but it’s really dense and has a lot of technical language that is not common knowledge for some people who have expertise in things like alignment and evals, but not working in your narrow niche. (like me)
Also, I would love to see more information on how your findings connect with the broader things that AI safety people should care about.
I know that making research is hard, and hope that my comments are helpful.