Tags Discussion/Talk Thread

I find the "A / B" to be fairly ugly in tag naming, and think that even "A (and B)" is more attractive.

My guess is that we should just go with Bayesianism, because it feels more general? Like, if it's a wiki page later, the page on Bayesianism would naturally have a section on Bayes' Theorem that explains it.

FWIW I vote for "Bayes' Theorem" over "Bayes Theorem".

(I'm pro renaming for now.)

Just had a chance to look at these. Thanks!!

I really like what you've done with Self-Deception and Superstimuli. They're both really solid.

Courage is good too, though I find it slightly hard to read. "here" references seem less good than just using the post name.

Evolution I found myself bouncing off a bit for some reason. Gave it C-Class for now, but I still hope to revise the grading-scheme/description-improvement-system following all the useful feedback/advice you gave. I'm just getting back on top of things following our team-retreat, and next priorities are getting drawn up now. Similarly, we need to decide about doing a push to get tag a basic description, which was also a good thing you suggested.

I also left a comment suggesting this. for now i'm just adding "Related Sequences" to the description with links to relevant sequences (see Epistemology [? · GW] for example), i hope in the future this can be done with actual tagging.

Yeah, I thought about this early on and it seemed very necessary. It's not lost on me though how it's kinda bad to have 20 posts from the same sequences just filling up the tag.

As you point out, there area bunch of challenges in figuring out all the right behaviors. There might be a solution where the benefits are worth it, but it's gonna be work to figure out what it is. We're first dealing with a heap of low-hanging fruit on tagging stuff.

In short, seems like a good idea to figure something out here, also seems tricky and so would be a bit further down the roadmap. For now, I think the Related Sequences is the best solution even if some people miss the section.

Merge Blues and Greens [? · GW] and Coalitional Instincts [? · GW]; they're about basically the same thing. I don't like either name; "Tribalism" would probably be better. Blues and Greens is jargon that's not used enough, and coalitional instincts is too formal.

I don't have an opinion on the Blues and Greens merge - I wouldn't expect anyone to be specifically interested in posts that happen to use that particular analogy - but I would somewhat lean towards keeping the Coalitional Instincts term.

I considered several terms for that tag, including "Tribalism", but I feel like there's an underlying concept cluster that's worth carving out and which is better described by Coalitional Instincts. Though this feels like a somewhat subtle difference in what I feel that "Tribalism" connotes, and if others disagree with me on those connotations, then I'm certainly willing to switch.

Basically, it feels to me like "Tribalism" generally refers to a somewhat narrow set of behaviors, whereas Coalitional Instincts includes those but also includes the underlying psychological mechanisms and somewhat broader behaviors. For example, Eliezer's post Professing and Cheering [LW · GW] includes this excerpt:

But even the concept of “religious profession” doesn’t seem to cover the pagan woman’s claim to believe in the primordial cow. If you had to profess a religious belief to satisfy a priest, or satisfy a co-religionist—heck, to satisfy your own self-image as a religious person—you would have to pretend to believe much more convincingly than this woman was doing. As she recited her tale of the primordial cow, she wasn’t even trying to be persuasive on that front—wasn’t even trying to convince us that she took her own religion seriously. I think that’s the part that so took me aback. I know people who believe they believe ridiculous things, but when they profess them, they’ll spend much more effort to convince themselves that they take their beliefs seriously.

It finally occurred to me that this woman wasn’t trying to convince us or even convince herself. Her recitation of the creation story wasn’t about the creation of the world at all. Rather, by launching into a five-minute diatribe about the primordial cow, she was cheering for paganism, like holding up a banner at a football game. A banner saying Go Blues isn’t a statement of fact, or an attempt to persuade; it doesn’t have to be convincing—it’s a cheer.

That strange flaunting pride . . . it was like she was marching naked in a gay pride parade.1 [LW(p) · GW(p)]It wasn’t just a cheer, like marching, but an outrageous cheer, like marching naked—believing that she couldn’t be arrested or criticized, because she was doing it for her pride parade.

That’s why it mattered to her that what she was saying was beyond ridiculous. If she’d tried to make it sound more plausible, it would have been like putting on clothes.

To me, "believing in something because it is outrageous, and not even trying to make it legible according to the other people's epistemology" is a phenomenon that's covered by coalitional instincts - it's holding up the standards of explanation of your group and flaunting the fact that you don't care about the other side's. But it doesn't quite fit tribalism as I usually understand it? I feel like "tribalism" usually refers to things that are more explicitly about "explicitly attacking anything that is seen to support the other side", and not so much about "simply being proud about your side".

But I'm not completely certain of that myself, and if others disagree with this assessment, then I'd be willing to change the name to tribalism.

I like Blues and Greens being separate, reserved for posts using that specific analogy, as opposed to other posts on the topic the analogy bears on. The analogy is flavorful, and it's made its way into our jargon.

Yeah, those seem like sensible merges to me. Habryka mentioned something about wanting to do a big merge/clean-up in a couple of weeks. We might also want to build tools to make merging/splitting for the occasion.

I've pinged the four tag creators now, though, in case they want to argue in particular directions now.

(I'm still getting a handle on exactly what an A-grade tag should feel like though, honestly).

Me too. Maybe I'll get a chance to write some of my own today or tomorrow and we can compare notes.

Probably the hard thing is that what counts as A-Class for one posts isn't for enough for another. For Heuristics & Biases, man, it's such a central tag for LW so it's gotta be really good. Probably we should hold off declaring this one A-Class until we're ready to say it really doesn't merit more work, which could be a while.

One thought is that this a topic with some amazing posts introducing it and the tag description should lean on them, maybe even quoting them a lot and pushing them to read them. The current tag description is alright, but it feels like it doesn't get at the heart of the topic in the few paragraphs or make me feel like I should care much. Contrast with the engagingness of ...What's a bias, again? [LW · GW]

In Conservation of Expected Evidence [? · GW], I basically thought to myself "there's no way I'm gonna write a better explanation than Eliezer on this", should just quote him. Though sometime's Eliezer's explanations are too long to quote and it makes sense to rewrite them. 

Other thoughts:
- I think posts in this class should heavily mention and describe top posts close to the top (including pretty related ones like "Your Intuitions Aren't Magic"'). 
- Relatedly, it should guide your reading of the topic much more explicitly. What is Predictably Wrong? What does it contain?
- I think it'd be good to have a more explicit list of different heuristics and biases. As kind of a parent tag, it should maybe even have a nice table of all "sub-tags"
- The way "fallacies" is used on LW isn't about explicit logic, really. I think it's more that a fallacy is a bad inference/step of reasoning, whereas heuristics and biases are properties of the algorithm that does the reasoning. 

[I also moved the Related Tags to the top because I think it's good if someone can immediately see whether a related tag is what they're after (even in the hover-over which you can hackily include if you shift-enter)]

I missed this comment till now, sorry. Seems no one else opined yet, so my thoughts:

I don't feel like I can confidently say something about the value of distinguishing the two, but I can offer the broader questions I see.

1. What is the extent of our ability to maintain nuanced distinctions in the tags?

I've seen already that tag creators will have one narrow thing in mind when they make a tag, and then other people will come along and apply it very broadly, e.g. Biology getting applied to anything involving a biological system at all even if it'd be of no interest to a biologist or someone look for biologist content. If two tags are very adjacent, I might expect things to haphazardly go in one or the other or both.

      2. If it's important, how do we maintain two tags?

I could see preserving a space to just discuss the formal side separately from the psychological being quite valuable. Especially if there's a sustained back-and-forth on technical stuff. If it is, then I think there's effort we could put in to maintain the distinction in the face of entropy.

• The tags need to be optimized a set.
• The tag names directly should imply the tag boundary, e.g Truth (formal) and Map & Territory (Applied) or something.
• Each description should early on differentiate the tags and mention the other.
• You'll need someone who cares to keep an eye on the tags. (Subscriptions to tags should make this easier, but currently, it doesn't batch that doesn't work well for high-volume tags).
• Although if each tag has highly upvoted relevant content at the top, it matters less if some other stuff and goes to the bottom of the list (especially below the Load More)

Overall, the seeming cost of maintaining separate tags makes me by default want to just aim for broader, more-inclusive tags. But if a distinction is worth it and/or we end up with a committed community and people who want to keep tags clean and true to their intention, might be worth it.

I'd say that the appropriate number of tags for a post is "as many or few as seem to match the contents of the post".

I expect that many posts that have just a few tags are that way simply because the tagger didn't happen to think of the other possible ones. Or someone might be focusing on some particular tag and tagging all relevant posts with that, without also looking at those posts to see what other tags might be suitable.

I put 1-3 on most posts, but I've gone up to 5 or more on some. Probably many of the posts I've tagged could have other tags applied to them that I didn't think of at the time. It's not about a hard number, it's about asking for each individual tag, is it likely someone exploring this tag would think this post was relevant to it?

I'm glad you asked!

I feel a bit uncomfortable with "piling on tags" (maybe the phrase more than what you're actually suggesting). I think it's because when I've seen authors apply lots of tags to their own posts (I'm assuming the motivation), half the tags seem like a stretch and the posts were low karma/quality (less true of yours, but it's an association now in my mind).

That said, I do think 4-5 tags can be reasonable, and more posts might hit that or more as we have more tags and people do more tagging. I'm mostly responding to the phrasing "piling on for easy visibility". The heuristics Kaj and Multicore suggest feel right to me.

If the relevance is really there, it's good to tag.

Out of these I think Dissolving the Question is probably the right name for the tag. Dissolving the Cognitive Algorithm is in interesting alternate name for the technique, but since it isn't known it's not very good for the tag name. How an algorithm feels from the inside doesn't feel like a tag name, and wouldn't be intuitive to put on posts that aim to dissolve questions.

Though Dissolving the Question also feels awkward for a tag name. Perhaps 'Dissolving Questions", of if anyone has better ideas?

I've felt like I've wanted a tag for "Confusions" and I guess by extension one about "Deconfusions"? "Deconfusioning?"

I'm really not sure, but I think there's a broader thing here that should a be a (the?) tag.  Something about the broader phenomenon of getting deconfused, of which "dissolving the question" is an instance.

Of the three things listed tough, definitely Dissolving the Question.

Phenomenal job on the EA tag! Definitely A-Class. 

Quick thoughts that occur to me for it: Should probably link to the EA Forum too? And I think it might be clearer if the See Also section was divided into two sub-sections, one for related LW tags and one for external resources?

I made Epistemic Modesty C-Class. The central idea is conveyed but I'd want to reserve B-Class for things that touch on more of the arguments, e.g., for and against. Inadequate Equilibria posts are tagged, but would be good to mention what position gets laid out in them. (Especially as the quote might make it seem that Eliezer is plain in favor of epistemic modesty as virtuous.)

PS: See a note on tag feedback in this comment [LW(p) · GW(p)].

I think AI Risk is open to improvement as a name, but it's definitely a more narrow category than AI. AI includes reviews of AI textbooks, explanation of how certain ML architectures work, and just anything relating to AI. AI risk is about the downside risk and analysis of what the risk looks like.

I think we should definitely not tag every chapter of HPMOR. So to be consistent we should probably untag the later chapters of 3 Worlds and make that the norm. I guess that should be added to the description.

Thanks for the very generous offer though!

I believe yes, though we haven't gotten around to fully deprecating/redirecting, etc.

I created a "Case Study" tag which seems to overlap with both of these a bit?  Definitely seems like it could be different.

After reading the Arbital page on Mild Optimization, it seems like a distinct cluster so I'm going ahead and making a new tag. And adding a link to that page from the Mild Optimization tag.

Also thanks for your prodigious tagging effort! Let me know if you want to chat (send you a DM).

I'd use the machine learning tag [? · GW].