There's a striking contrast between the Yudkowsky of 2019 who wrote the "Reducing Negativity" post, and an earlier Yudkowsky (from even before the Sequences) who maintained [a page on Crocker's rules](http://sl4.org/crocker.html): if you declare that you operate under Crocker's rules, you're consenting to other people optimizing their speech for conveying information rather than being nice to you. If someone calls you an idiot, that's not an "insult"; they're just informing you about the fact that you're an idiot, and you should probably thank them for the tip. (If you _were_ an idiot, wouldn't you be better off knowing that?)
-It's of course important to stress that Crocker's rules are opt in on the part of the receiver; it's not a license to unilaterally be rude to other people. Adopting Crocker's rules as a community-level norm on an open web forum does not seem like it would end well.
+It's of course important to stress that Crocker's rules are opt-in on the part of the receiver; it's not a license to unilaterally be rude to other people. Adopting Crocker's rules as a community-level norm on an open web forum does not seem like it would end well.
Still, there's something precious about a culture where people appreciate the obvious normative ideal underlying Crocker's rules, even if social animals can't reliably live up to the normative ideal. Speech is for conveying information. People can say things—even things about me or my work—not as a command, or as a reward or punishment, but just to establish a correspondence between words and the world: a map that reflects a territory.
-Appreciation of this obvious normative ideal seems strikingly absent from Yudkowsky's modern work—as if he's given up on the idea that reasoning in public is useful or possible.
+Appreciation of this obvious normative ideal seems strikingly absent from Yudkowsky's modern work—as if he's given up on the idea that reasoning in public is useful or possible. His [_Less Wrong_ commenting guidelines](/images/yudkowsky_commenting_guidelines.png) declare, "If it looks like it would be unhedonic to spend time interacting with you, I will ban you from commenting on my posts." The idea that people who are unhedonic to interact with might have intellectually substantive criticisms that the author has a [duty to address](https://www.greaterwrong.com/posts/pC74aJyCRgns6atzu/meta-discussion-from-circling-as-cousin-to-rationality/comment/kS4BfYJuZ8ZcwuwfB) does not seem to have crossed his mind.
The "Reducing Negativity" post also warns against the failure mode of attempted "author telepathy": attributing bad motives to authors and treating those attributions as fact without accounting for uncertainty or distinguishing observations from inferences. I should be explicit, then: when I say negative things about Yudkowsky's state of mind, like it's "as if he's given up on the idea that reasoning in public is useful or possible", that's a probabilistic inference, not a certain observation.
-But I think making probabilistic inferences is ... fine? The sentence "Credibly helpful unsolicited criticism should be delivered in private" sure does look to me like text generated by a state of mind that doesn't believe that reasoning in public is useful or possible.[^criticism-inference] I think that someone who did believe in public reason would have noticed that criticism has information content whose public benefits might outweigh its potential to harm an author's reputation or feelings.[^unhedonic] If you think I'm getting this inference wrong, feel free to let me _and other readers_ know why in the comments.
-
-[^criticism-inference]: More formally, I'm claiming that the [likelihood ratio](https://arbital.com/p/likelihood_ratio/) P(wrote that sentence|doesn't believe in public reason)/P(wrote that sentence|does believe in public reason) is greater than one.
-
-[^unhedonic]: Speaking of authors' feelings, it's notable that Yudkowsky's [_Less Wrong_ commenting guidelines](/images/yudkowsky_commenting_guidelines.png) declare, "If it looks like it would be unhedonic to spend time interacting with you, I will ban you from commenting on my posts." The text does not seem to consider the possibility that people who are unhedonic to interact with might have intellectually substantive contributions to make.
+But I think making probabilistic inferences is ... fine? The sentence "Credibly helpful unsolicited criticism should be delivered in private" sure does look to me like text generated by a state of mind that doesn't believe that reasoning in public is useful or possible. I think that someone who did believe in public reason would have noticed that criticism has information content whose public benefits might outweigh its potential to harm an author's reputation or feelings. If you think I'm getting this inference wrong, feel free to let me _and other readers_ know why in the comments.
-----
That was my intuition. To formalize it, I wanted some sensible numerical quantity that would be maximized by using "nice" categories and get trashed by gerrymandering. [Mutual information](https://en.wikipedia.org/wiki/Mutual_information) was the obvious first guess, but that wasn't it, because mutual information lacks a "topology", a notion of "closeness" that would make some false predictions better than others by virtue of being "close".
-Suppose the outcome space of _X_ is `{H, T}` and the outcome space of _Y_ is `{1, 2, 3, 4, 5, 6, 7, 8}`. I wanted to say that if observing _X_=`H` concentrates _Y_'s probability mass on `{1, 2, 3}`, that's more useful than if it concentrates _Y_ on `{1, 5, 8}`. But that would require the numerals in _Y_ to be numbers rather than opaque labels; as far as elementary information theory was concerned, mapping eight states to three states reduced the entropy from lg<sub>2</sub> 8 = 3 to lg<sub>2</sub> 3 ≈ 1.58 no matter which three states they were.
+Suppose the outcome space of _X_ is `{H, T}` and the outcome space of _Y_ is `{1, 2, 3, 4, 5, 6, 7, 8}`. I wanted to say that if observing _X_=`H` concentrates _Y_'s probability mass on `{1, 2, 3}`, that's more useful than if it concentrates _Y_ on `{1, 5, 8}`. But that would require the numerals in _Y_ to be numbers rather than opaque labels; as far as elementary information theory was concerned, mapping eight states to three states reduced the entropy from log<sub>2</sub> 8 = 3 to log<sub>2</sub> 3 ≈ 1.58 no matter which three states they were.
How could I make this rigorous? Did I want to be talking about the variance of my features conditional on category membership? Was "connectedness" what I wanted, or was it only important because it cut down the number of possibilities? (There are 8!/(6!2!) = 28 ways to choose two elements from `{1..8}`, but only 7 ways to choose two contiguous elements.) I thought connectedness was intrinsically important, because we didn't just want _few_ things, we wanted things that are similar enough to make similar decisions about.
projects / Ultimately_Untrue_Thought.git / blobdiff