About a year ago, LessWrong held its second annual review, where we looked over the best posts from 2019. The LessWrong team offered $2000 in prizes for the top post authors, and (up to) $2000 in prizes for the best reviews of those posts.

Once again, some things came up and we got pretty sidetracked in actually getting around to awarding those prizes. Apologies to authors and reviewers. (I'm setting in motion some processes so that this year's prizes are given out in a much more timely fashion)

Prizes for Posts

The $2000 in prizes for posts this year are awarded to:

$200 to Paul Christiano, for What Failure Looks Like
$600 (divided evenly) among Evan Hubinger, Chris van Merwijk, vlad_m, Joar Skalse and Scott Garrabrant, for Risks from Learned Optimization
$200 to Abram Demski, for Parable of the Predict-o-matic and Selection vs Control
$200 to John Wentworth, for Being on the Pareto Frontier
$200 to Scott Alexander, for Book Review: Secret of our Success and Rule Thinkers in, not Out
$200 to Kaj Sotala, for Unlocking Emotional Brain
$200 to Zvi Mowshowitz, for Asymmetric Justice
$200 to Zack Davis, for Heads I Win, Tails I Never Heard of Her

The above authors are chosen primarily by the outcome of the vote (on which they all scored very highly). I'm also making a moderator call to award $100 to Benquo for Reason Isn't Magic, essentially coming out of our review-prize-budget. I thought it was a good review of the Secrets of Our Success post, which I think warranted special attention (see below).

Prizes for Reviews

Reviewing generally seems unglamorous compared to authoring posts. To compensate,I awarded more prize money here per unit-of-effort than I did for posts.

I particularly went out of my way to reward reviews that attempted a factual epistemic spot check – not because I think those are most important overall, but because they seem undersupplied to me, and more effortful. (AFAICT we only had 2-3?)

That said, much of LessWrong's content is in the domain of philosophy or theory, where epistemic spot checks aren't quite the right frame.

Secrets of Our Success

Book Review: Secrets of Our Success was an interesting case that stood out to me when reflecting on last year's process. Fiddler's review basically argues "This post mostly quotes cherry picked anecdotes, without distinguishing which of those anecdotes Scott endorses, and some of those cherry picked anecdotes don't even check out." No one rebuts Fiddler's review.

This seems worthy of investigation. I talked to some people about how they felt about Secrets of Success. A few people I respect said they agreed with some criticisms, and nonetheless voted for it because it contributed an important frame for looking at the world.

I'm still mulling over how I personally think about the concept. Meanwhile, it seemed valuable for our community reward signal to highlight critical discussion of the post, since it was rated as the #4 post in the review vote.

The awards for review prizes are:

$500 to John Wentworth both for writing several in-depth, valuable reviews, as well as being very comprehensive in his reviews. Notable shout out for his review of Coherent Decisions Imply Consistent Utilities, and Seeking Power is Often Convergently Instrumental in MDPs.
$300 to Zvi for writing a large set of reviews and nominations, covering a number of topics, demonstrating useful judgment on each of them.
$200 to Adam Shimi for several in-depth reviews on alignment related topics.
$200 to Turntrout for his review on But Exactly How Complex and Fragile, as well as the surrounding discussion.
$200 to Fiddler for his review of Book Review: Secret of Our Success, as well as several other nuanced reviews.
$100 to Rohin for engaging in debate in John's review of Coherent Decisions Imply Consistent Utilities.
$100 to AllAmericanBreakfast for his review of Reason Isn't Magic, among other things actually looking into what's up with Manioc plant and how it works.
$100 to Abram Demski for his review of Asymmetric Justice.
$100 to Magfrump for his epistemic spot check on From Personal To Prison Gangs. John had some interesting responses about whether that spot check was modeling the situation usefully. I felt like I learned things about how to reason about spot-checks from the exchange.

If he weren't on the Lightcone team and ineligible for prizes, I'd be awarding jacobjacob a significant prize both for his many uses of the Elicit tool to crowdsource evaluation of claims during the Review. I also appreciated his epistemic spot check on Approval Extraction Advertised As Production – I disagreed with his method of evaluation, but liked the effort at an epistemic spot check, which I think was undersupplied in the Review. I also liked how jimrandomh came in with an alternate evaluation.

Congratulations to all prize winners!

I'm super grateful to all the authors both for the original work that went into their posts, and their willingness to engage with a longterm feedback system, and to the reviewers for providing that feedback. There's plenty of improvement to build into our annual review system, but I'm enthusiastic about us continuing to build towards a robust intellectual ecosystem.

I'll be PMing you all shortly about how to claim your winnings.