So I think it is likely that they don’t release a regular product anytime soon and instead are going to just come out with ASI randomly (with some USG involvement).
I hope they share their alignment strategy before they randomly release ASI. As I recall, Ilya had some ideas around scalable oversight which didn’t seem super promising, but maybe the plans have changed or they’ve made further progress since then.
I mainly want to see what strategy they used out of curiosity. I don’t think they are just running RLHF on whatever deep-learning based super efficient learner novel architecture they cooked up. Would be interesting to see!
If there isn’t a superefficient novel architecture, then Sutskever is to be arresred for wholesale fraud. If there is, then Sutskever’s startup makes the world LESS safe to live in, and Sutskever failed at bringing about his promises. How could one arrange for the potential architecture to be leaked into Anthropic (or, at least, OAI/GDM) for thorough analysis?
I hope they share their alignment strategy before they randomly release ASI. As I recall, Ilya had some ideas around scalable oversight which didn’t seem super promising, but maybe the plans have changed or they’ve made further progress since then.
I mainly want to see what strategy they used out of curiosity. I don’t think they are just running RLHF on whatever deep-learning based super efficient learner novel architecture they cooked up. Would be interesting to see!
If there isn’t a superefficient novel architecture, then Sutskever is to be arresred for wholesale fraud. If there is, then Sutskever’s startup makes the world LESS safe to live in, and Sutskever failed at bringing about his promises. How could one arrange for the potential architecture to be leaked into Anthropic (or, at least, OAI/GDM) for thorough analysis?