A month ago, the LessWrong team launched our new tagging system [LW · GW]...with a whopping whole entire single tag [? · GW]. We rushed it out quickly to give people a way to manage the deluge of Covid-19 posts which had grown to 50% of new posts by mid-March.
Now we’re ready to trial the tag system generally with all kinds of tags. With this launch people can now 1) use tag pages to find posts about their interests, and 2) filter their Latest Posts section on the frontpage for the content they’re most interested in. Soon tag filtering will applicable to Recent Discussion and All Posts page, as well as it being possible to subscribe to tags.
How experimental is this?
Very. The team is figuring out all kinds of questions related to tag ontology and tagging UI, and it’s all subject to change. We’d greatly appreciate feedback and user interviews (comment on this post or use the usual channels [LW · GW]!)
Warning! For the next while, if you help us by tagging posts or voting on tag relevance, it’s possible you will lose your effort if we later decide to delete the tag due to our evolving tag policy. We appreciate anyone who tags anyway since this helps us develop the tagging system.
For example, we've created the World Optimization high-level core tag for everything which is about concretely improving the world. But the name isn't great– it doesn't imply that personal improvement is included. So we need to figure out a better name or maybe split the tag up.
Click on the tag filter gear icon next to Latest Posts.
Note the option for “soft filtering” which lets you titrate the amount you see posts with a given tag from Neverthrough Normal Amountto Exclusively. (We’re still working to make this UI intuitive.)
Click on a Tag on the bottom of a post page to visit its Tag Page and find related posts.
Visit the Tag Index (Beta) page to see the list of high-quality existing tags.
Who can tag things?
Anyone can tag a post with an existing tag.
Ideally tags have a definition paragraph which explain when they do and don’t apply, but as we ramp up not all posts will contain this.
Who can vote on tags?
Anyone can vote on how relevant a tag is to a given post. This helps order posts on a tag page. Users’ “tag relevance” voting power is the same as their normal karma voting power See Quick Guide to Tagging#Tag-Relevance [LW · GW] for more detail.
Who can create tags?
Right now, only LessWrong moderators can create tags. If you really want a particular tag to exist, comment here or contact us via Intercom, email, or our FB page. See Quick Guide to Tagging#Propose-tags [LW · GW] for criteria for good tag proposals.
Why isn’t this in beta?
Usually it would have been, but since we rolled out the beginnings of the tag system already, we decided to launch this unfolding experiment directly to everyone.
What are the Core Tags?
We’ve decided upon six tags that cover the most important categories which LessWrong posts typically fall under. We hope to soon ensure that these tags have complete coverage of all posts on LessWrong (accurately tagged for all posts).
LessWrong has the goal of causing intellectual progress on important problems [LW · GW]. That requires being more than a “news site” where people read posts when they’re published and then a week later those posts are forgotten. Rather, people need to able to find the valuable posts of yesteryear so they can learn from them and build upon them in steadily growing, “shoulders-of-giants”-style accumulation of communal knowledge.
Tagging, along with search, wiki, and pingbacks/citations, are key tools for ensuring past content remains easily findable to those who ought to find it. Tagging enables the following:
People who read a post they like on a topic can easily find other posts on that topic.
A researcher catching up on the “conversation” in an ongoing research area can get up to date more quickly by finding the most important posts for a concept.
People can filter their frontpage/All Post experiences to show them the content they want to see, increasing the individuals Signal-to-Noise ration on the site.
People can subscribe to tags to stay up to date on topics of their interests [coming].
It’s possible that major tags might even get something of a "subreddit treatment” and have their own discussion sections or at least Open Threads.
Yes please! Comment here or contact us via the usual channels (Intercom, email, questions [? · GW], FB) if you’ve got thoughts or are willing to do a user interview via Skype/Zoom.
What do you think of the UI so far? Which tags do you really want? Which use-cases and features matter most to you? Or tell us about the tag systems you love from elsewhere. More generally, tell us how you like to find content you value on LessWrong and elsewhere.
Taking a look at the Tag Index [LW · GW], I think Betting [? · GW] would be the closest existing tag, but it's not a great fit. And what I'm thinking of should probably go under the Individual Optimization section of World Optimization in the hierarchy, rather than under Rationality.
(It might also make sense to create tags for Money or Career that would also fall under Individual Optimization, to catch posts like Maximizing Your Donations via a Job [LW · GW], but I haven't thought as much about those as tags.)
These are great! I'll make these soon. Those posts definitely justify doing so in my mind. Re: Wei Dai's comment, I think it's reasonable to mention in the tag description text (and those will soon be everyone-editable wiki entries and should include extra info relevant to the tag/wiki/concept, including "notable comments").
Here are some maybe useful tags. Interpret these as ideas, not requests.
Mechanism Design (I think I am imagining including systemization that aligns incentives within yourself in here, which maybe means you would want a more general name like "Aligning Incentives" but I think I prefer "Mechanism Design")
Fake Frameworks (When I first thought of this, I was thinking of people tagging their own posts. Maybe it is a little weird to have people tagging each other's posts as fake. )
Embedded Agency (Where I am imagining this as being largely for technical work) (In particular, I personally would get more use out one big embedded agency tag than a bunch of smaller tags, since I feel like all the most interesting stuff in embedded agency cuts across tags like "decision theory")
Something like the class including: Toward a New Technical Explanation of Technical Explanation, Embedded World Models, technical logical uncertainty work, things about dealing with the fact that Bayes is not a viable strategy for embedded agents. "Embedded World Models" "Resource Bounded Epistemics" "Embedded Epistemics" "Post-Bayesianism" I would hope the name here does not make people think it should only be for technical things.
Something like the class including: How I Lost 100 Pounds Using TDT, Humans Are Embedded Agents Too, Inner alignment in the brain, Sources of intuitions and data on AGI, things about applying AI alignment theory to human rationality and vice versa. Maybe more generally about applying results from one field to another field. "Interdisciplinary Analogies"?
Mechanism Design/Aligning Incentives seems good too. Agree there are choices about the name, and I guess scope too. Do you mean it to be material about how to align incentives but exclude related stuff of examples where incentives failed to be aligned. Would Boeing 737 MAX MCAS as an agent corrigibility failure [LW · GW] be part of it?
"Resource Bounded Epistemics" sounds like a cool category. So does "Interdisciplinary Analogies", or should it be "Interdisciplinary Applications"?
Anyhow, these are great. More are welcome.
Fake Frameworks, yeah, hmm. We might consider "only authors can apply these tags", I'm not sure. Those might make sense for general "epistemic state" tags.
+1 for a Mechanism Design/Aligning Incentives tag. I think "incentive design" would be a good name for this category. This would encompass material on specification gaming, tampering, impact measures, etc. Including specific examples of misaligned incentives under this umbrella seems fine as well.
Is the "aligning incentives" tag you are interested in something AI specific or should it apply to general human institutions / social systems? I could see a case for either, but that impacts what tag names we should use.
I have now created a Philosophy of Language tag. I haven't yet created a "disagreement" tag because it feels like it could use a more precise name. "Philosophy of Disagreement" is... okay but not great.
Both sound reasonable at first glance. I'm not quite sure if "disagreement" is sufficient as tag-name (name way "language" feels a bit underspecified but "philosophy of language" disambiguates pretty well).
It'd be useful if the search function allowed searching for tags, as that'd likely be quicker than clicking through the tags page. A synonym feature would probably also be useful so that someone could try tag it with X and that would be replaced by the canonical tag Y instead.
I'd suggest meta-rationality as it's own core tag, but I imagine that'd be controversial.
Perhaps it might be easier to have a hierarchy of tags - so that voting for Value learning also votes for AI Alignment say
I have a question related to this -- I just went to find That Alien Message [LW · GW] to mark it with the AI Alignment [? · GW] tag, but then found it was already marked with the AI Boxing (Containment) [? · GW] tag. If Boxing is meant to be a subtag under the AI Alignment category, then I would think there's nothing more to do, but when I go to the Boxing page, there doesn't seem to be anything explicitly linking it to the AI Alignment page.
And That Alien Message doesn't show up as one of the posts in the list on the AI Alignment page, but I'd think it should, since I consider it part of the AI safety "canon" as one of the best sources for building a particular intuition.
So, should That Alien Message be explicitly tagged with AI Alignment? Or Should AI Boxing (Containment) be explicitly linked to AI Alignment somehow? Or does how I'm thinking about how all this should work differ from what the LW team is thinking?
Yeah, I really don't know yet how we should handle hierarchical tagging. My current model is that "That Alien Message" should just be tagged both AI Alignment and AI Boxing, and that we should be hesitant to have too much formal hierarchy in the tags, since I expect that most relationships will turn out to be only "almost hierarchical" in that it's not literally always the case the we would want to apply both tags together, and being forced to always have them come in a package is I think pretty costly.
I do think the description of the AI Boxing tag should link to the AI Alignment tag, so that should be edited in. And in general related tags should link to each other a lot, to make it easier to discover content that way.
I do think there will be an appreciable number of tags (even if they're a minority) that are strictly subsets of, say, AI alignment. Like everything under Value Learning or Embedded Agents, etc, and maybe it's worth it to have that automatically update.
I do feel tag descriptions linking to other tags is extremely important for the system to work and will help a lot here.
Have you considered requiring posters to apply at least one of the main five tags before their post can go up? This could reduce mod overhead and get people more involved in their own tagging. However, it is a small barrier to posting.
If you did this, it would probably be wise to display the main five tags by default instead of just putting them in the search box, or having the search box automatically recommend the main five tags in the drop-down before you search anything.
To untag a post, just downvote its tag relevance. (Either in the hover-over or on the tag page).
Yeah, agree with need a better solution for showing currently available tags. In the meantime, you can look at www.lesswrong.com/tags or www.lesswrong.com/tags/all
A heuristic the team has discussed is that tags should have 3 good post by at least two different authors. I do want some kind of wellbeing category, and a separate health one make sense too. Anatomy, if it isn't a topic discussed by others, may or may not make sense. I'm not sure. If it's to help people find your other writing (the main goal of tagging), you could create a sequence or two to link them.
I suggest decoupling “modeling” and “optimization” from “world.” In fact, what does “world” even mean there? Usecase in mind: An “equilibria” tag for posts about modeling and understanding equilibria and posts about changing (optimizing) them. Currently this needs two new tags, but with decouplement it would be just one.
recommendations: posts where the author recommends something to others. (Like the “Connected papers” post)
recommendation-aggregation: posts where the author asks for recommendations. (Like the “best textbooks” post.)
transient: For posts whose relevance is highly time sensitive. A lot of covid posts fall under this, for example. This can help people avoid reading stale posts in the future. Ideally, we want a system where we can filter out transient posts from X time ago.
controversial: These posts should be shown only to users with some karma threshold.
platitude: posts whose content is not novel but the message merits repetition.
empirical: posts that analyze empirical data. A lot of SSC’s posts fall under this category.
technical-math: posts that require mathematical maturity to understand (i.e., beyond a good high school math foundation)
technical-cs: posts that require familiarity with CS and programming to understand.
help-needed: posts that need some help to accomplish their objective. E.g., someone has a good software idea but needs help in implementing it. The tag is used a lot on Github, but I think it can be used for things other than software, too.
One of the main reason I was starting Facebook groups was so that people interested in just one specific topic could see a list about this topic. So I think this will definitely be useful! Although Facebook groups / pages might still be useful so that people see LW posts in their newsfeed. I wrote more about this idea on Facebook.
Have you considered using Telegram groups+channels? A Telegram channel can have an associated group. The importance stuff go into the channel, which is auto-forwarded to the group, and people can discuss stuff in the group. The network effects are against Telegram, true, but it is quite a superior platform. It also doesn’t perpetuate the bad design of most social media that waste a lot of your time on ultimately unimportant stuff.
Interesting. I agree we want more specific tags too for that post. Though "Problem-Solving Tactics", actually feels pretty broad too, though a good definition/description might help give it shape. I'll think about it, not sure if you had one in mind.
Another thing that helps is having other posts in mind too for the tag.
I can't think of other posts, but the idea is a class of general tactics that can be applied to difficult technical problems, such as Polya's heuristics for solving math problems: generalization, specialization, using special cases, pattern-matching to related problems, etc. There are others:
Use fresh eyes
Prove a stronger statement than needed
When there are many candidate strategies, use manual depth-first search and keep track efficiently
The tag wiki for the "techniques" tag pointed out by habryka says "set of actions (including "mental actions") for improving one's thinking so as to form accurate beliefs and/or make better decisions", which I think is slightly different, but the tag wiki could be edited to encompass problem-solving tactics.
"Rationality" is part of a special class of 5 tags that are designed to be general, and be primarily used for broad filtering purposes (i.e. the 5 core tags [LW · GW]). Agree that it would be far too broad for most tags.
Re. "Problem-Solving Tactics": I agree that something in that space is good. We currently have the "Techniques" tag, which is pretty similar to that, so I added it to that tag. Do you think we want something again more specific than that?