Conflux

When Are Two Net­works the Same? Ten­sor Similar­ity for Mechanis­tic Interpretability

• Conflux 12 Nov 2024 6:24 UTC

  −1 points

  0

  on: o1 is a bad idea

  I appreciated this! I didn’t know much about o1, and this gives me a much better understanding of how it’s different. My brain finds Abram very trustworthy for some reason.

• Conflux 8 Aug 2024 22:36 UTC

  64 points

  16

  on: How I Learned To Stop Trust­ing Pre­dic­tion Mar­kets and Love the Arbitrage

  Hi! I’m the person that wrote that Manifold newsletter. My opinion: you’re totally right, and it was sloppy to include that as justification that Whitmer was a reasonable choice.

  In the counterfactual universe where that conditional market didn’t exist, I might have written that section anyway; it started because I personally believed that a Harris-Whitmer ticket would have about those odds, and it just so happened that the conditional market agreed so I cited it. (In my head it was fine because Whitmer had some chance to be the VP in the past, so the odds couldn’t be totally garbage, but any sense of internal ranking among the contenders was pretty silly.) I’m glad I gave it some disclaimer, but I think it wasn’t a strong enough disclaimer.

  As mentioned below, I think conditional markets aren’t always useless; especially those with a >25% chance of occurring and where the correlation-causation issues aren’t too severe. But I’m embarrassed Manifold was promoting this one so much, and I’ll try to fight to make the site more intellectually honest going forward.

  Thanks for raising this issue! (…and Whitmer’s probability)

• Conflux 10 Jul 2024 17:47 UTC

  3 points

  2

  on: AirBnB Baking

  Your comment about flax eggs is very validating to me; I used them once, to make vegan brownies when I did the Seven-Day Vegan Challenge, and they were not good. I’ve now updated substantially in favor of trying them for recipes like muffins!

• Conflux 28 Dec 2021 18:50 UTC

  3 points

  0

  on: Merry Christmas

  I don’t think I’ve ever commented on one of your posts, but I’ve probably read about fifty of them! Seeing your name listed as the author of a post is a good guarantee that I’ll find it interesting and thought-provoking.

• Conflux 14 Jun 2021 1:48 UTC

  3 points

  0

  on: Bad names make you open the box

  “Regression to the mean” is also known as “reversion to the mean,” by the way, which I think is a clearer name.

• Conflux 25 Mar 2021 0:15 UTC

  5 points

  0

  on: A vot­ing the­ory primer for rationalists

  I’m quite interested in voting systems, but I was surprised to discover that the general consensus is that score beats approval! I checked it out and it seems to be a robust finding that in real life people understand & are happier with score, but this surprised me.

  I’d think that since there are so many options for score, it’d be a bit overwhelming and hard to figure out how to optimize. Whereas with approval it’s basically “vote for the minor candidates you like better than the major ones; and also vote for your least unfavorite major candidate.” Which is simple enough.

  I can more clearly see the argument for three to five scoring options instead of approval. If you’re, say, a Warren supporter in the 2020 election (which of course is now a free-for-all approval voting bonanza) do you vote for Biden or not? If you don’t vote for him, you risk throwing the election to Trump; whereas if you do, you may end up giving the election to Biden over Warren. So a middle-ground-y thing seems reasonable here. (A counterargument here is that with polling, it can be more clearly seen whether a mass effort by the Democratic Party Establishment to get liberals in line is necessary, or whether people can coalesce behind Warren or Sanders or whoever.)

  Three options nicely correlates with “like, neutral, dislike” and five nicely with “love, like, neutral, dislike, hate” as heuristics for honest voting. I’m just apprehensive about a hundred. I don’t even know how I’d vote with more than three candidates to rate on such a high scale, I shudder to think how a low-information voter would do it.

• Conflux 23 Nov 2020 21:55 UTC

  1 point

  0

  on: Norms of Mem­ber­ship for Vol­un­tary Groups

  I thought this provided a lot of clarity about membership norms! I definitely have had the experience of arguing with people based on different understandings of which norms the server/​group operated under.

  To me, the textbook example of something that runs under Coalition Norms is a for-profit business. Does this not fit in that category cleanly, an oversight, or just omitted to keep the list of examples minimalistic?

• Conflux 20 Oct 2020 1:48 UTC

  2 points

  0

  on: It’s hard to use util­ity max­i­miza­tion to jus­tify cre­at­ing new sen­tient beings

  I think I have a pretty simple solution to this: treat 0 as the point when each being is neither happy nor unhappy. Negative numbers are fine. You can still take the sum. In the example, this seems like just subtracting 10 from everyone, which is 0 in the dogless state and −3 with the dog. Thus: no puppy.