Curated! I think this post is a considerable contribution to the ecosystem and one that many people are grateful for. Progress is made by people building on the works of others, and for that to happen, people have to be aware of the works of others and able to locate those most relevant to them. As the Alignment field grows, it gets progressively harder to keep up with what everyone is up to, what’s been tried, where more effort might be useful. Roundups like these enable more people to get a sense of what’s happening much more cheaply. And seeing an overview all at once helps distill the bigger picture and questions.
In my case, a thing in this roundup that wasn’t previously salient to me is that it’s hard to find modularity in neural nets. I feel a bit indignant about that. Why should that be hard? I feel incline to poke at the problem and think about it, and who knows, maybe contribute some progress.
But thinking again about this review. A I’m aware has been raised is how to keep something like this accurate and up to date. That feels like something LessWrong arguably should try to do, and we do have the wiki-tag system, so how to build a system of tech + people that does this work is something this post prompts me to think about. Kudos!
Curated! I think this post is a considerable contribution to the ecosystem and one that many people are grateful for. Progress is made by people building on the works of others, and for that to happen, people have to be aware of the works of others and able to locate those most relevant to them. As the Alignment field grows, it gets progressively harder to keep up with what everyone is up to, what’s been tried, where more effort might be useful. Roundups like these enable more people to get a sense of what’s happening much more cheaply. And seeing an overview all at once helps distill the bigger picture and questions.
In my case, a thing in this roundup that wasn’t previously salient to me is that it’s hard to find modularity in neural nets. I feel a bit indignant about that. Why should that be hard? I feel incline to poke at the problem and think about it, and who knows, maybe contribute some progress.
But thinking again about this review. A I’m aware has been raised is how to keep something like this accurate and up to date. That feels like something LessWrong arguably should try to do, and we do have the wiki-tag system, so how to build a system of tech + people that does this work is something this post prompts me to think about. Kudos!