Posts
Comments
The list of once “secret” documents is very cool, thanks for that. (But I skimmed the other parts too)
I think the interchangeability is just hard to understand. Even though I know they are the same thing, it is still really hard to intuitively see them as being equal. I personally try (but not very hard) to stick with X -> Y in mathy discussions and if/only if for normal discussions
For nondeterministic voting surely you can just estimate the expected utility of your vote and decide whether voting is worth the effort. Probably even easier than deterministic ones.
Btw, I feel like the post is too incomplete on its own for the title Should we abstain from voting?. It feels more like Why being uninformed isn't a reason to not vote.
Maybe make a habit of blocking https://www.lesswrong.com/posts/*
while writing?
The clickbait title is misleading, but I forgive this one because I do end up finding it interesting, and it is short. In general I mostly don't try to punish things if it end up to be good/correct.
Starting today I am going to collect a list of tricks that websites use to prevent you from copy and pasting text + how to circumvent them. In general, using ublock origin and allow right click properly fixes most issues.
1. Using href (https://lnk.to/LACA-15863s, archive)
behavior: https://streamable.com/sxeblz
solution: use remove-attr in ublock origin - lnk.to##.header__link:remove-attr(href)
2. Using a background image to cover the text (https://varium.jp/talent/ahiru/, archive)
Note: this example is probably just incompetence.
behavior: https://streamable.com/bw2wlv
solution: block the image with ublock origin
3. Using draggable=true (Spotify album titles, archive)
Note: Spotify does have a legit reason to use draggable. You can drag albums or tracks to your library, but I personally prefer having texts to be selectable.
behavior: https://streamable.com/cm0t6b
solution: use remove-attr in ublock origin - open.spotify.com##.encore-internal-color-text-base.encore-text-headline-large.encore-text:upward(1):remove-attr(draggable)
4. Using EventListener
s to nullify selections (https://www.uta-net.com/song/2765/, archive)
behavior: https://streamable.com/2i1e9k
solution: ublock origin seem to just work, but you can probably try intercepting addEventListener to nullify the nullifier
To answer your question directly - not really.
I think index pages are just meant to be used by only a small minority of people in any community. In my mind, the LW concepts page is like the wiki topic groups (not sure what they're called).
The similarities are:
- It is fun to go through the concepts page and find tags I haven't learned about, this is good for exploration but a rare use case (for me)
- Because it is an index, it is useful when you have a concept in your mind but couldn't remember the name
But the concepts page has a worse UX than wiki since you have to explicitly search for it, rather than it popping up in the relevant tags page, and also they show up in a cluster
How do you use them?
I use it when I am interested in learning about a specific topic. I rarely use the Concepts page, because it contains too many tags, and sometimes I don't even know what tag I am looking for. Instead, I usually already have one or two articles that I have previously read, which feels similar to the topic I am thinking about. I would then search for those posts, look at the tags, and click on the one that is relevant. In the tag page, I start by reading the wiki, but often feel disappointed by the half-done/incompleteness of the wiki. Then I filter by high karma and read the articles from top to bottom, skipping ones that feels irrelevant or uninteresting based on title.
Do you wish you could get value from them better?
I wish the default most relevant ordering is not based on the raw score, but rather a normalized relevance score or something more complicated, because right now it means nothing other that "this post is popular so a lot of people voted on the tags". This default is really bad, every new user has to independently realize that they should change the sorting. LW also does not remember the sorting so I have to change it manually every time, which is irritating but not a big deal.
I understand that having the audio player above the title is the path of least resistance, since you can't assume there is enough space on the right to put it in. But ideally things like this should be dynamic, and only take up vertical space if you can't put it on the right, no? (but I'm not a frontend dev)
Alternatively, I would consider moving them vertically above the title a slight improvement. It is not great either, but at least the reason for having the gap is more obvious.
The above screenshots are done in a 1920x1080 monitor
I like most of the changes, but strongly dislike the large gap before the title. (I similarly dislike the large background in the top 50 of the year posts)
Seems like every new post - no matter the karma - is getting the "listen to this post" button now. I love it.
I do believe that projects in general often fail due to lack of glue responsibilities, but didn't want to generalize too much in what I wrote.
Start with integration. Get the end-to-end WORKING MOCKUP going with hardcoded behaviors in each module, but working interfaces. This is often half or more of the work, and there's no way to avoid it - doing it at the end is painful and often fails. Doing it up front is painful but actually leads to completion.
Being able to convince everyone to put in the time to do this upfront is already a challenge :/ Sometimes I feel quite hopeless?/sad? in that I couldn't realistically make some coordination techniques work because of everyone's difference of goals and hidden motivations, or the large upfront cost in building a new consensus away from the Schelling point of normal university projects.
A common failure mode in group projects is that students will break up the work into non-overlapping parts, and proceed to stop giving a fuck about other's work afterwards because it is not their job anymore.
This especially causes problems at the final stage where they need to combine the work and make a coherent piece out of it.
- No one is responsible for merging the work
- Lack of mutual communication during the process means that the work pieces cannot be nicely connected without a lot of modifications (which no one is responsible for).
At this point the deadline is likely just a couple days (or hours) away, everyone is tired of this crap and don't want to work on it, but the combined work is still a piece of incoherent crap.
I wonder how I can do better at coordination while dealing with normal peers and while only doing a fair amount of work.
Thanks for adding a much more detailed/factual context! This added more concrete evidence to my mental model of "ELO is not very accurate in multiple ways" too. I did already know some of the inaccuracies in how I presented it, but I wanted to write something rather than nothing, and converting vague intuitions into words is difficult.
Take with a grain of salt.
Observation:
- Chess engines during development only play against themselves, so they use a relative ELO system that is detached from the FIDE ELO. https://github.com/official-stockfish/Stockfish/wiki/Regression-Tests#normalized-elo-progression https://training.lczero.org/?full_elo=1 https://nextchessmove.com/dev-builds/sf14
- It is very hard to find chess engines confidently telling you what their FIDE ELO is.
Interpretation / Guess: Modern chess engines probably need to use like some intermediate engines to transitively calculate their ELO. (Engine A is 200 ELO greater than players at 2200, Engine B is again 200 ELO better than A...) This is expensive to calculate and the error bar likely increases as you use more intermediate engines.
I follow chess engines very casually as a hobby. Trying to calibrate chess engine's computer against computer ELO with human ELO is a real problem. I doubt extrapolating IQ over 300 will provide accurate predictions.
Ranting about LangChain, a python library for building stuff on top of llm calls.
LangChain is a horrible pile of abstractions. There are many ways of doing the same thing. Every single function has a lot of gotchas (that doesn't even get mentioned in documentations). Common usage patterns are hidden behind unintuitive, hard to find locations (callbacks has to be implemented as an instance of a certain class in a config TypedDict). Community support is non-existent despite large number of users. Exceptions are often incredibly unhelpful with unreadable stack trace. Lots of stuff are impossible to type check because langchain allows for too much flexibility, they take in prompt templates as format strings (i.e. "strings with {variables}") and then allows you to fill in the template at runtime with a dict, so now nothing can be statically type checked :)
There are a few things I dislike about math textbooks and pdfs in general. For example, how math textbooks often use theorems that are from many pages ago and require switching back and forth. (Sometimes there isn't even a hyperlink!). I also don't like how proofs sometimes go way too deep into individual steps and sometimes being way too brief.
I wish something like this exists (Claude generated it for me, prompt: https://pastebin.com/Gnis891p)
Many people don't seem to know when and how to invalidate the cached thoughts they have. I noticed an instance of being unable to cache invalidate the model of a person from my dad. He is probably still modelling >50% of me as who I am >5 years ago.
The Intelligent Social Web briefly talked about this for other reasons.
A lot of (but not all) people get a strong hit of this when they go back to visit their family. If you move away and then make new friends and sort of become a new person (!), you might at first think this is just who you are now. But then you visit your parents… and suddenly you feel and act a lot like you did before you moved away. You might even try to hold onto this “new you” with them… and they might respond to what they see as strange behavior by trying to nudge you into acting “normal”: ignoring surprising things you say, changing the topic to something familiar, starting an old fight, etc.
In most cases, I don’t think this is malice. It’s just that they need the scene to work. They don’t know how to interact with this “new you”, so they tug on their connection with you to pull you back into a role they recognize. If that fails, then they have to redefine who they are in relation to you — which often (but not always) happens eventually.
I would like the option to separate subscribing to posts and subscribing to comments. I mostly just want to subscribe to posts, because it is much easier to decide whether I want to read a post than a comment.
that is much clearer that I think you should have said it out loud in the post
I also mostly switched to browser bookmark now, but I do think even this simple implementation of in-site bookmarks is overall good. Book marking in-site can sync over devices by default, and provides more integrated information.
I want to be able to quickly see whether I have bookmarked a post to avoid clicking into it (hence I suggested it to be a badge, rather than a button like in the Bookmarks tab). Especially with the new recommendation system that resurfaces old posts, I sometimes accidentally click on posts that I bookmarked months before.
This is like raw, n=1, personal feedback.
No, not really. I read it twice but couldn't bring myself to care. It seems you are going into tangents and not actually talking directly about your technique. I could be wrong, but I also couldn't care enough to read into the sentences and understand what you're actually pointing at with all the words. Having conclusion is nice because I jumped straight to that at first, seems kind of too normal to justify the clickbait though. Overall I feel like I read some ramblings and didn't learn much.
I would suggest using less clickbaity titles on LessWrong
I would love to get a little bookmark symbol on the frontpage
Metaphor rebranded themselves. No and no, thanks for sharing though, will try it out!
Related: Replace yourself before you stop organizing your community.
I think this is a important skill to learn and a important failure mode to be aware of. There are so many nice things that are gone because the only person that is doing the thing just stopped one day.
Unplugging the charger and putting it in my bag (and the reverse) is a trivial inconvenience that annoyed me many times
I found that it is possible to yield noticeably better search results than Google by using Kagi as default and fallback to Exa (prev. Metaphor).
Kagi is $10/mo though with a 100 searches trial. Kagi's default results are slightly better than Google, and it also offers customization of results which I haven't seen in other search engines.
Exa is free, it uses embeddings and empirically it understood semantics way better than other search engines and provide very unique search results.
If you are interested in experimenting you can find more search engines in https://www.searchenginemap.com/ and https://github.com/The-Osint-Toolbox/Search-Engines
You can buy a even cheaper one from taobao, this one is $20 before shipping (but I expect the buying experience to be quite complex if you're outside china)
I pattern matched onto "10 Times Scientists Admitted They Were Wrong" and thought you made some sort of editing mistake, now I see what you're saying with the added quotes
The title is messed up
I subscribed to your posts and got an in-site notification
(My native language is Chinese.) I haven't started reading, but I am finding the abstract/tldr impossible to understand. For example, "Is the accuracy higher between forecasts" reads like a nonsensical sentence. My best guess after reading one extra paragraph by click through is that the question is actually "are forecasts predicting the near future more accurate than those predicting a more distant future" but I don't feel like it is possible to decode just based on the abstract. I have similar issues with all three questions.
I noticed this too, but when trying to rank music based on my taste. I wonder if when people are asked to give their favorite (of something), do they just randomly give a maximal element, or do they have an implicit aggregate function that kind of converts the partial order into a total order
EigenKarma is a (basically abandoned) attempt to apply a semi-transitive trust algorithm in real situations. EigenTrust (the original paper, which is the basis of EigenKarma) had some brief discussions and computational experiments on how adversaries affect trust.
I had been wanting to spend some time working on improving EigenKarma for the last year but haven't got around to do it.
Github is known to occasionally be unreachable in China. In general I think people in LW should be able to figure out VPNs
I just stumbled on this website: https://notes.andymatuschak.org/About_these_notes It has a similar UI but for Obsidian-like linked notes. The UI seem pretty good.
Sorry, I don't feel like completely understanding your POV is worth the time. But I did read you reply 2-3 times. In roughly the same order as your writing.
Yes, so if you observe no sabotage, then you do update about the existence of a fifth column that would have, with some probability, sabotaged (an infinite possibility). But you don't update about the existence of the fifth column that doesn't sabotage, or wouldn't have sabotaged YET, which are also infinite possibilities.
I'm not sure why infinity matters here, many things have infinite possibilities (like any continuous random variable) and you can still apply a rough estimate on the probability distribution.
I guess it's a general failure of Bayesian reasoning. You can't update 1 confidence beliefs, you can't update 0 confidence beliefs, and you can't update undefined beliefs.
I think this is an argument similar to an infinite recursion of where do priors come from? But Bayesian updates usually produces better estimate than your prior (and always better than your prior if you can do perfect updates, but that's impossible), and you can use many methods to guestimate a prior distribution.
You have a pretty good model about what might cause the sun to rise tomorrow, but no idea, complete uncertainty (not 0 with certainty nor 1 with certainty, nor 50/50 uncertainty, just completely undefined certainty) about what would make the sun NOT rise tomorrow, so you can't (rationally) Bayesian reason about it. You can bet on it, but you can't rationally believe about it.
Unknown Unknowns are indeed a thing. You can't completely rationally Bayesian reason about it, and that doesn't mean you can't try to Bayesian reason about it. Eliezer didn't say you can become a perfect Bayesian reasoner either, he always said you can attempt to reason better, and strive to approach Bayesian reasoning.
Relatedly, in-line private feedback. I saw a really good design for alerting typos here.
To the four people who picked 37 and thought there was a 5% chance other people would also choose it, well played.
Wow, that's really a replicable phenomenon
Threads are pretty good, most help channels should probably be a forum (or 1 forum + 1 channel). Discord threads do have a significant drawback of lowering visibility by a lot, and people don't like to write things that nobody ever sees.
^ Forum
I didn't read either links, but you can write whatever you want on LessWrong! While most posts you see are very high quality, this is because there is a distinction between frontpage posts (promoted by mods) and personal blogposts (the default). See Site Guide: Personal Blogposts vs Frontpage Posts.
And yes some people do publish blogposts on LessWrong, jefftk being one that I follow.
FAQ: What can I post on LessWrong?
Posts on practically any topic are welcomed on LessWrong. I (and others on the team) feel it is important that members are able to “bring their entire selves” to LessWrong and are able to share all their thoughts, ideas, and experiences without fearing whether they are “on topic” for LessWrong. Rationality is not restricted to only specific domains of one’s life and neither should LessWrong be. [...]
I tend to think of "keep my identity small" as "keep my attachments to identity dimensions weak".
Very much agree.
suggestions:
- Duplicate this to the open thread to increase visibility
- I don't know your exact implementation for forming the ranked list, but I worry that if you (for example) simply sort from low likelihood to high likelihood, it encourages people to only submit very low probability predictions.
For possible solutions:
1. This is my problem and I should find a way to stop feeling ugh
2. Have some ways to easily read a summary of long comments (AI or author generated)
3. People should write shorter comments on average
I often have an ugh feeling towards reading long comments.
Posts are usually well written, but long comments are usually rambly, even the highest karma ones. It takes a lot of effort to read the comments on top of reading the post, and the payoff is often small.
But for multiple reasons, I still feel an obligation to read at least some comments, and ugh.
You'd need perhaps 100, maybe even 1,000 times more arguments to get a perfectly open-minded and Bayesian agent to start from the point where the other person started and end up agreeing with you.
Modelling humans with Bayesian agent seems wrong.
For humans, I think the problem usually isn't the number of arguments / number of angles you attacked the problem, but whether you have hit on the few significant cruxes of that person. This is especially because humans are quite far away from perfect Bayesians. For relatively small disargreements (i.e. not at the scale of convincing a Christian that God doesn't exist), usually people just had a few wrong assumptions or cached thoughts. If you can accurately hit those cruxes, then you can convince them. It is very very hard to know which arguments can hit those cruxes though and it is why one of the viable strategies is to keep throwing arguments until one of them work.
(Also unlike convincing Bayesian agents where you can argue for W->X, X->Y, Y->Z in any order, sometimes you need to argue about things in the correct order)