This is a minor point, but I am somewhat worried that the idea of research debt/research distillation seems to be getting diluted over time. The original article (which this post links to) says:
Distillation is also hard. It’s tempting to think of explaining an idea as just putting a layer of polish on it, but good explanations often involve transforming the idea. This kind of refinement of an idea can take just as much effort and deep understanding as the initial discovery.
I think the kind of cleanup and polish that is encouraged by the review process is insufficient to qualify as distillation (I know this post didn’t use the word “distillation”, but it does talk about research debt, and distillation is presented as the solution to debt in the original article), and to adequately deal with research debt.
There seems to be a pattern where a term is introduced first in a strong form, then it accumulates a lot of positive connotations, and that causes people to stretch the term to use it for things that don’t quite qualify. I’m not confident that is what is happening here (it’s hard to tell what happens in people’s heads), but from the outside it’s a bit worrying.
I actually made a similar comment a while ago about a different term.
Personally, I believe I understood “research debt” in the strong way upon first reading (I hadn’t encountered the term before, but the post included a definition), but was then immediately struck by the inadequacy of the review process to address the problem. Granted, it’s a move in the right direction.
(note: this view isn’t necessarily shared by other LW team members, and there’s at least some kind of major disagreements about how to think about this)
In my ideal world, the review process outputs “here’s what’s inadequate about all our best posts”, and there’s a period where authors are expected to actually make substantial improvements before they go in the book. This is a lot to ask of authors who are often pretty busy, and I’m not 100% sure it’s the right way to go about things but I lean that way.
Alternately, there might be a LessWrong Textbook that is “the level above the Review”, where the Review takes stock of which posts are ready for being officially canonized (which has quite high standards), but that’s pretty rare.
I am somewhat confused about how to think about things like… say, Simulacra, where there’s already been ongoing efforts to distill/refine/clarify. Benquo might choose to rewrite his original post in response to review-feedback, but Zvi has sort of already been working on that (but, on later posts that wouldn’t show up till the 2020 review)
I think the distillation function is better served by new posts, as opposed to extensive revisions of old posts. The repeated attempts to re-explain simulacra are a good example of this.
Part of why I think this is because I believe editing old posts, especially extensive editing, loses some “aliveness” from the post which lets me get in touch with the author’s thought process as they write. Let the rough off the cuff posts stay rough.
Another reason is that textbooks and high quality survey papers are not edited versions of old papers.
“Radical Probabilism” was in many ways a more mature version of “toward a new technical explanation of technical explanation”, but could never be produced by editing the old post.
Benquo might choose to rewrite his original post in response to review-feedback, but Zvi has sort of already been working on that (but, on later posts that wouldn’t show up till the 2020 review)
If Benquo re-wrote his original post extensively enough, that would also sort of count as 2020-2021 content. Which makes me wonder whether it would make sense to distinguish ‘this post is in the 2018 review because of its underlying content’ vs. ‘this post is in the 2018 review because of its implementation/presentation’? Then the ‘underlying content’ stuff could include multiple posts, or posts from later years, as warranted.
A bad situation to maybe try to avoid is: an important idea never ends up included in a review because the original exposition is in one post/year, and the best exposition is in another post/year, and the original exposition is not quite well-executed to warrant inclusion, while the best exposition is not quite noteworthy or innovative enough. (E.g., maybe the best exposition is just a shorter, lightly rephrased version of the original.)
Including content and not just posts also helps address the problem where you want to credit multiple different people for an idea (or idea+exposition), even where their collaboration was spread across multiple posts rather than a single post with multiple authors.
Yeah agreed. There’s (relatedly) the thing where “post” isn’t even always the natural category of thing-people-want-to-nominate (often I think people are nominating a sequence). It’s a somewhat tricky question “how do we let people nominate ‘concepts’ with multiple related posts?” in a way that has good UI and is clear.
I think for immediate future, it’s probably good to just manually spell out “I’m really nominating this overall concept, and am interested in comparing it to more recent work”. I’m not 100% sure what’ll make sense for handling that in the Review Phase but seems worth trying.
Yes, sorry. The concrete mechanism by which I hope to address research debt is not the editing of essays, but the identification of essays that have good ideas and bad presentation, and encouraging other authors to write better new explanations for them, as well as more something like thorough rewrites of existing posts.
I see, that wasn’t clear from the post. In that case I am wondering if the 2018 review caused anyone to write better explanations or rewrite the existing posts. (It seems like the LessWrong 2018 Book just included the original posts without much rewriting, at least based on scanning the table of contents.)
At least 3 people substantially rewrote their posts in the 2018 review, and my hope is that over time it becomes pretty common for there to be substantial rewriting. (albeit, two of those people were LessWrong team members)
But for what it’s worth, here’s the diff between the original version of my own post and the current version I wrote as a result of the review.
One of the pernicious issues with word-dillution is that often when people try to use a word to mean things, they’re… kinda meaning those things “aspirationally.” Where, yes part of my original goal with the Review absolutely included Research Debt. But indeed there’s a decent chance it won’t succeed at that goal. (But, I do intend to put in a fair amount of optimization pressure towards making it succeed)
This is a minor point, but I am somewhat worried that the idea of research debt/research distillation seems to be getting diluted over time. The original article (which this post links to) says:
I think the kind of cleanup and polish that is encouraged by the review process is insufficient to qualify as distillation (I know this post didn’t use the word “distillation”, but it does talk about research debt, and distillation is presented as the solution to debt in the original article), and to adequately deal with research debt.
There seems to be a pattern where a term is introduced first in a strong form, then it accumulates a lot of positive connotations, and that causes people to stretch the term to use it for things that don’t quite qualify. I’m not confident that is what is happening here (it’s hard to tell what happens in people’s heads), but from the outside it’s a bit worrying.
I actually made a similar comment a while ago about a different term.
Personally, I believe I understood “research debt” in the strong way upon first reading (I hadn’t encountered the term before, but the post included a definition), but was then immediately struck by the inadequacy of the review process to address the problem. Granted, it’s a move in the right direction.
(note: this view isn’t necessarily shared by other LW team members, and there’s at least some kind of major disagreements about how to think about this)
In my ideal world, the review process outputs “here’s what’s inadequate about all our best posts”, and there’s a period where authors are expected to actually make substantial improvements before they go in the book. This is a lot to ask of authors who are often pretty busy, and I’m not 100% sure it’s the right way to go about things but I lean that way.
Alternately, there might be a LessWrong Textbook that is “the level above the Review”, where the Review takes stock of which posts are ready for being officially canonized (which has quite high standards), but that’s pretty rare.
I am somewhat confused about how to think about things like… say, Simulacra, where there’s already been ongoing efforts to distill/refine/clarify. Benquo might choose to rewrite his original post in response to review-feedback, but Zvi has sort of already been working on that (but, on later posts that wouldn’t show up till the 2020 review)
I think the distillation function is better served by new posts, as opposed to extensive revisions of old posts. The repeated attempts to re-explain simulacra are a good example of this.
Part of why I think this is because I believe editing old posts, especially extensive editing, loses some “aliveness” from the post which lets me get in touch with the author’s thought process as they write. Let the rough off the cuff posts stay rough.
Another reason is that textbooks and high quality survey papers are not edited versions of old papers.
“Radical Probabilism” was in many ways a more mature version of “toward a new technical explanation of technical explanation”, but could never be produced by editing the old post.
If Benquo re-wrote his original post extensively enough, that would also sort of count as 2020-2021 content. Which makes me wonder whether it would make sense to distinguish ‘this post is in the 2018 review because of its underlying content’ vs. ‘this post is in the 2018 review because of its implementation/presentation’? Then the ‘underlying content’ stuff could include multiple posts, or posts from later years, as warranted.
A bad situation to maybe try to avoid is: an important idea never ends up included in a review because the original exposition is in one post/year, and the best exposition is in another post/year, and the original exposition is not quite well-executed to warrant inclusion, while the best exposition is not quite noteworthy or innovative enough. (E.g., maybe the best exposition is just a shorter, lightly rephrased version of the original.)
Including content and not just posts also helps address the problem where you want to credit multiple different people for an idea (or idea+exposition), even where their collaboration was spread across multiple posts rather than a single post with multiple authors.
Yeah agreed. There’s (relatedly) the thing where “post” isn’t even always the natural category of thing-people-want-to-nominate (often I think people are nominating a sequence). It’s a somewhat tricky question “how do we let people nominate ‘concepts’ with multiple related posts?” in a way that has good UI and is clear.
I think for immediate future, it’s probably good to just manually spell out “I’m really nominating this overall concept, and am interested in comparing it to more recent work”. I’m not 100% sure what’ll make sense for handling that in the Review Phase but seems worth trying.
I would be excited for the review to output “Here’s a description of further work we’d like to see done”.
Yes, sorry. The concrete mechanism by which I hope to address research debt is not the editing of essays, but the identification of essays that have good ideas and bad presentation, and encouraging other authors to write better new explanations for them, as well as more something like thorough rewrites of existing posts.
I see, that wasn’t clear from the post. In that case I am wondering if the 2018 review caused anyone to write better explanations or rewrite the existing posts. (It seems like the LessWrong 2018 Book just included the original posts without much rewriting, at least based on scanning the table of contents.)
At least 3 people substantially rewrote their posts in the 2018 review, and my hope is that over time it becomes pretty common for there to be substantial rewriting. (albeit, two of those people were LessWrong team members)
But for what it’s worth, here’s the diff between the original version of my own post and the current version I wrote as a result of the review.
Thanks! That does make me feel a bit better about the annual reviews.
One of the pernicious issues with word-dillution is that often when people try to use a word to mean things, they’re… kinda meaning those things “aspirationally.” Where, yes part of my original goal with the Review absolutely included Research Debt. But indeed there’s a decent chance it won’t succeed at that goal. (But, I do intend to put in a fair amount of optimization pressure towards making it succeed)