Bluntly, I don’t really understand what you take issue with in my review. From my perspective the structure of my review goes like this:
We are currently in an alignment winter. (This is bad)
Alignment is not solved yet but people widely believe it is. (This is bad)
I was expecting to hate the book but it actually retreats on most of the rhetoric I blame for contributing to the alignment winter. (This is good)
The style of the book is bad, but I won’t dwell on it and in fact spend a paragraph on the issue and then move on.
I actually disagree with the overall thesis, but think it’s virtuous to focus on the points of agreement when someone points out an important issue so I don’t dwell on that either and instead
“Emphatically agree” (literal words) that AI labs are not serious about the alignment problem.
State a short version of what the alignment problem actually is. (Important because it’s usually conflated with or confused with simpler problems that sound a lot easier to solve.)
I signal boost Eliezer’s other and better writing because I think my audience is disproportionately made up of people who might be able to contribute to the alignment problem if they’re not deeply confused about it and I think Eliezer’s earlier work is under-read.
I reiterate that I think the book is kinda bad, since I need a concluding paragraph.
I continue to think this is a basically fair review.
I had to reread part 7 from your review to fully understand what you were trying to say. It’s not easy to parse on a quick read, so I’m guessing Zvi didn’t interpret the context and content correctly, like I didn’t on my first pass. On first skim, I thought it was a technical argument about how you disagreed with the overall thesis, which makes things pretty confusing.
Bluntly, I don’t really understand what you take issue with in my review. From my perspective the structure of my review goes like this:
We are currently in an alignment winter. (This is bad)
Alignment is not solved yet but people widely believe it is. (This is bad)
I was expecting to hate the book but it actually retreats on most of the rhetoric I blame for contributing to the alignment winter. (This is good)
The style of the book is bad, but I won’t dwell on it and in fact spend a paragraph on the issue and then move on.
I actually disagree with the overall thesis, but think it’s virtuous to focus on the points of agreement when someone points out an important issue so I don’t dwell on that either and instead
“Emphatically agree” (literal words) that AI labs are not serious about the alignment problem.
State a short version of what the alignment problem actually is. (Important because it’s usually conflated with or confused with simpler problems that sound a lot easier to solve.)
I signal boost Eliezer’s other and better writing because I think my audience is disproportionately made up of people who might be able to contribute to the alignment problem if they’re not deeply confused about it and I think Eliezer’s earlier work is under-read.
I reiterate that I think the book is kinda bad, since I need a concluding paragraph.
I continue to think this is a basically fair review.
I had to reread part 7 from your review to fully understand what you were trying to say. It’s not easy to parse on a quick read, so I’m guessing Zvi didn’t interpret the context and content correctly, like I didn’t on my first pass. On first skim, I thought it was a technical argument about how you disagreed with the overall thesis, which makes things pretty confusing.