The LessWrong 2019 Review

Today is the start of the 2019 Review, continuing our tradition of checking which things that were written on LessWrong still hold up a year later, and to help build an ongoing canon of the most important insights developed here on LessWrong.

The whole process will span 8 weeks, starting on December 1st:

  • From December 1st to the 14th, any user that was registered before January 1st 2019 can nominate any post written in 2019 to be considered for the review.

  • From December 14th to January 11th, any user can leave reviews on any posts with at least two nominations, ask questions of other users and the author, and make arguments for how a post should be voted on in the review.

  • From January 11th to January 25th any LessWrong user registered before 2019 can vote on the nominated posts, using a voting system based on quadratic voting. (There will be two votes, one for 1000+ karma users, and one for all users)

But before I get more into the details of the process, let’s go up a level.

Why run a review like this?

The Review has three primary goals:

  1. Improve our incentives, feedback, and rewards for contributing to LessWrong.

  2. Create a highly curated “Best of 2019” sequence and physical book

  3. Create common knowledge about the LW community’s collective epistemic state about the most important posts of 2019

Improving our incentives and rewards

Comments and upvotes are a really valuable tool for allocating attention on LessWrong, but they are ephemeral and frequently news-driven, with far-from-perfect correlation to the ultimate importance of an idea or an explanation.

I want LessWrong to be a place for Long Content. A place where we can build on ideas over decades, and an archive that helps us collectively navigate the jungle of infinite content that spews forward on LessWrong every year.

One way to do that is to take some time between when you first see a post and when you evaluate it. That’s why today we are starting the 2019 review, not the 2020 review. A year is probably enough time to no longer be swept away in the news or excitement of the day, but recent enough that we can still remember and write down how an idea or explanation has affected us.

I also want LessWrong to not be overwhelmed by research debt:

Research debt is the accumulation of missing interpretive labor. It’s extremely natural for young ideas to go through a stage of debt, like early prototypes in engineering. The problem is that we often stop at that point. Young ideas aren’t ending points for us to put in a paper and abandon. When we let things stop there the debt piles up. It becomes harder to understand and build on each other’s work and the field fragments.

There needs to be an incentive to clean up ideas that turned out to be important but badly presented. This is the time for authors to get feedback on which of their posts turned out to be important and to correct minor errors, clean up prose and polish them up. And the time for others to see what concepts still lack a good explanation after at least a whole year has passed, and to maybe take the time to finally write that good canonical reference post.

Creating a highly curated sequence and book

The internet is not great at preserving things for the future. Also, books feel real to me in a way that feels very hard to achieve for a website. Also, they look beautiful:

One of the books printed for the 2018 Review.

Of course, when you show up to LessWrong, you can read Rationality: A-Z, you can read The Codex, and you can read HPMoR, but historically we haven’t done a great job at archiving and curating the best content of anyone who isn’t Scott or Eliezer (and even for Scott and Eliezer, it’s hard to find any curation of the content they wrote in recent years). When I showed up, I wish there was a best of 2012 book and sequence that would have helped me find the best content from the years from before I was active (and maybe we should run a “10-year Review” so that I can figure out what the best posts from 2010 and beyond are).

Create common knowledge

Ray says it pretty well in last year’s Review announcement post:

Some posts are highly upvoted because everyone agrees they’re true and important. Other posts are upvoted because they’re more like exciting hypotheses. There’s a lot of disagreement about which claims are actually true, but that disagreement is crudely measured in comments from a vocal minority.

Now is the time to give your opinions much more detail, distinguish between a post being an interesting hypothesis versus a robust argument, and generally help others understand what you think, so that we can discover exciting new disagreements and build much more robustly on past and future work.

What does it look like concretely?


Nominations really don’t have to be very fancy. Some concrete examples from last year:

Reading Alex Zhu’s Paul agenda FAQ was the first time I felt like I understood Paul’s agenda in its entirety as opposed to only understanding individual bits and pieces. I think this FAQ was a major contributing factor in me eventually coming to work on Paul’s agenda. – evhub on “Paul’s research agenda FAQ”


This post not only made me understand the relevant positions better, but the two different perspectives on thinking about motivation have remained with me in general. (I often find the Harris one more useful, which is interesting by itself since he had been sold to me as “the guy who doesn’t really understand philosophy”.)

– Kaj Sotala on “Sam Harris and the Is-Ought Gap”

But sometimes can be a bit more substantial:

This post:

  • Tackles an important question. In particular, it seems quite valuable to me that someone who tries to build a platform for intellectual progress attempts to build their own concrete models of the domain and try to test those against history

  • It also has a spirit of empiricism and figuring things out yourself, rather than assuming that you can’t learning anything from something that isn’t an academic paper

  • Those are positive attributes and contribute to good epistemic norms on the margin. Yet at the same time, a culture of unchecked amateur research could end up in bad states, and reviews seem like a useful mechanism to protect against that

This makes this suitable for a nomination.

– jacobjacob on “How did academia ensure papers were correct in the early 20th Century?”

Overall, a nomination doesn’t need to require much effort. It’s also OK to just second someone else’s nomination (though do make sure to actually add a new top-level nomination comment, so we can properly count things).


We awarded $1500 in prizes for reviews last year. The reviews that we awarded the prizes for really exemplify what I hope reviews can be. The top prize went to Vanessa Kosoy, here’s an extract from one of her reviews

From Vanessa Kosoy on “Clarifying AI Alignment”:

In this essay Paul Christiano proposes a definition of “AI alignment” which is more narrow than other definitions that are often employed. Specifically, Paul suggests defining alignment in terms of the motivation of the agent (which should be, helping the user), rather than what the agent actually does. That is, as long as the agent “means well”, it is aligned, even if errors in its assumptions about the user’s preferences or about the world at large lead it to actions that are bad for the user.


In contrast, I will argue that the “motivation-competence” decomposition is not as useful as Paul and Rohin believe, and the “definition-optimization” decomposition is more useful.


The review both makes good arguments against the main thrust of the post it is reviewing, while also putting the article into a broader context that helps me place it in relation to other work in AI Alignment. She argues for an alternative breakdown of the problem where you instead of modeling it as the problems of “motivation and competence”, model it as the problems of “definition and optimization”. She connects both the decomposition proposed in the essay she is critiquing, and the one she proposed to existing research (including some of her own), and generally makes a point I am really glad to see surfaced during the review.

To be more concrete, this kind of ontology-level objection feels like one of the most valuable things to add during the review phase, even if you can’t propose any immediate alternative (i.e. reviews of “I don’t really like the concepts this post uses, it feels like reality is more neatly carved by modeling it this way” seem quite valuable and good to me).

Zack M. Davis was joint-second winner of prizes for reviews last year. Here’s an extract from a review of his.

Zack’s review “Firming Up Not-Lying Around Its Edge-Cases Is Less Broadly Useful Than One Might Initially Think”:

Reply to: Meta-Honesty: Firming Up Honesty Around Its Edge-Cases


A potential problem with this is that human natural language contains a lot of ambiguity. Words can be used in many ways depending on context. Even the specification “literally” in “literally false” is less useful than it initially appears when you consider that the way people ordinarily speak when they’re being truthful is actually pretty dense with metaphors that we typically don’t notice as metaphors because they’re common enough to be recognized legitimate uses that all fluent speakers will understand.


Zack wrote a whole post that I really liked, that made the core argument that while it might make sense to try really hard to figure out the edge-cases of lying, it seems that it’s probably better to focus on understanding the philosophy and principles behind reducing other forms of deception like using strategic ambiguity, heavily filtering evidence, or misleading metaphors.

Arguing that a post, while maybe making accurate statements, appears to put its emphasis in the wrong place, and encouraging action that seems far from the most effective, also seems generally valuable, and a good class of review.


You can trial-run the vote UI here (though you can’t submit any votes yet). Here is also a screenshot of what it looked like last year:

UI when first opening the review, during the basic voting pass
UI when using quadratic voting mode and selecting a post

How do I participate?

Now for some more concrete instructions on how to participate:


Starting today (December 1st), if you have an account that was registered before the 1st of January 2019, you will see a new button on all posts from 2019 that will allow you to nominate them for the 2019 review:

It’s at the top of the triple-dot menu.

Since a major goal of the review is to see which posts had a long-term effect on people, we are limiting nominations to users who signed up before 2019. If you were actively reading LessWrong before then, but never registered an account, you can ping me on Intercom (the small chat bubble in the bottom right corner on desktop devices), and I will give your account nomination and voting privileges.

I recommend using the All Posts page for convenience, where you can group posts by year, month, week, and day. Here’s the two I use the most:


Starting on December 14th, you can write reviews on any post that received more than 2 nominations. For the following month, my hope is that you read the posts carefully, write comments on them, and discuss:

  • How has this post been useful?

  • How does it connect to the broader intellectual landscape?

  • Is this post epistemically sound?

  • How could it be improved?

  • What further work would you like to see on top of the ideas proposed in this post?

I would consider the gold-standard for post reviews to be SlateStarCodex book reviews (though obviously shorter, since posts tend to be less long than books).

As an author, my hope is that you take this time to note where you disagree with the critiques, help other authors arrange followup work, and, if you have the time, update your post in response to the critiques (or just polish it up in general, if it seems like it has a good chance of ending up in the book).

This page will also you to see all posts above two nominations, and how many reviews they have, together with some fancy UI to help you navigate all the reviews and nominations that are coming in.


Starting on January 11th, any user who registered before 2019 can vote on any 2019 post that has received at least one review. The vote will use quadratic voting, with each participant having 500 points to distribute. To help handle the cognitive complexity of the quadratic voting, we also provide you with a more straightforward “No”, “Neutral”, “Good”, “Important”, “Crucial” scale that you can use to prepopulate your quadratic voting scores.

You can give the vote system a spin here on the posts from 2018, to get a sense of how it works and what the UI will look like.

Last year, only users above 1000 karma could participate in the review and vote. This year, we are going to break out the vote into two categories, one for users above a 1000 karma, and one for everyone. I am curious to see if and how they diverge. We might make some adjustments to how we aggregate the votes for the “everyone” category, like introducing some karma-weighting. Overall I expect we will give substantial prominence to both rankings, but favoring the 1000+ karma user ranking somewhat higher in our considerations for what to include in the final sequence and book. To be more concrete, I am imagining something like a 70:30 split of attention and prominence favoring the 1000+ karma users vote.

Prizes and Rewards

I think this review process is really important. To put the LessWrong’s Team’s money where it’s mouth is, we are again awarding $2000 in prizes to the top posts as judged by the review, and up to $2000 in prizes for the best reviews and nominations (as judged by the LW mod team). These are the nominations and reviews from last year that we awarded prizes.

Public Writeup and Aggregation

At the end of the vote, we are going to publish an analysis with all the vote results again.

Last year, we also produced an (according to me) really astonishingly beautiful book with all the top essays (thanks to Ben Pace and Jacob Lagerros!) and some of the best comments on reviews. I can’t promise we are going to spend quite as much time on the book this year, but I expect it to again be quite beautiful. See Ben’s post with more details about the books, and with the link to buy last year’s book if you want to get a visceral sense of them.

The book might look quite different for this year than it did for last year’s review, but still anyone who is featured in the book will get a copy of it. So even just writing a good comment can secure your legacy.

Good luck, think well, and have fun!

This year, just as we did last year, we are going to replace the “Recommendations & From the Archives” section of the site with a section that just shows you posts you haven’t read that were written in 2019.

I really enjoyed last year’s review, and am looking forward to an even greater review this year. May our epistemics pierce the heavens!