In computing, honeypots attract bad actors into interacting with a system. As usual, this principle can be reversed. You can deliberately put out some kind of intellectual honey that will attract _good_ actors instead, and lead to joyful serendipity. If you wear a shirt that says, “Dyslexics are teople poo”, that attracts one kind of conversation; a CERN standard model tee attracts another (in my case: “what’s it mean?”, “dunno”, the end.). The examples are endless – trail race swag attracts conversations with fellow trail runners; wearing something handmade gets the attention of other crafters – but conversations started by conversation pieces aren’t usually that deep or well-targeted. Conversation pieces are decent at starting conversations, but we don’t see that many people face-to-face (well, _I_ don’t!), so there’s a limit to how specific you can make the reference, and still meet people who get it. Henrik Karlsson (are you tired of me referencing Henrik Karlsson yet?) articulated the idea that a blog, or any public collection of work, is a way of passively collecting a network of interesting people that are unusually well-matched with your interests, without having to go to hundreds of speed-networking events in a nerdy tee shirt. A blog post is a very long and complex search query to find fascinating people and make them route interesting stuff to your inboxWhen writing in public, there is a common idea that you should make it accessible. This is a left over from mass media.Escaping FlatlandHenrik Karlsson This seems true. I’ve met a lot of interesting people by putting things out on the internet. But this, like most worthwhile things, can be fractal. First you go in public, then you wear a conversation piece, then you leave “handles” and “doorknobs” in your answers to the basic questions, creating a choose-your-own adventure for your conversation partner. First you make stuff, then you put it on the internet. If you stop here, you’re missing out! That’s like going out in public – with airpods in, and staring at the ground. The next step is to put lots of links to your contact form or email, and explicitly ask for feedback, and make it easy for people to say hello. Each post is a hook for a different conversation, with a different kind of person. Each post has a chance of pushing a reader over the activation energy threshold to start a conversation. Posts are, also, cumulative. Someone might read for years, and then write in to discuss draining pasta without a colander. (Other people might subscribe, and then realize, _oh, that’s not actually an $X blog, it’s a kitchen sink blog, and I’m only interested in 1/10th of the things they write about. Bye!_ If you have to pay to send emails, this is actually a good thing. They can still read your posts about $X when they bubble up in the $X communities they hang out in.) But within a blog, you can also place Easter eggs that target sub-segments of your readers. [ SPOILERS FOR WAITBUTWHY READERS AHEAD ] I’m old enough to remember the turtle in Wait Buy Why’s footer, which went to a page that said “You clicked the turtle”. This is a classic curiosity trap. The people who end up on that page are a small segment of their most engaged and curious readers. I don’t see the turtle in the footer anymore. _Only today_ did I discover that it’s turtles all the way down. The turtle page has a turtle that goes to another page, that has a smaller turtle that goes to another page that claims “You Win” but actually has a tiny turtle that goes to I actually can’t believe you, which has a giraffe that goes to Tim Urban’s old blog. Maybe it also has a pico-turtle that I missed. This was probably gave me the idea for using the interrobang as the icon for my /intractable page. And the whole idea of the intractable page is that it’s an _entire box of doorknobs_. Whether or not they found it via the tiny interrobang in the footer, nifty people have written in as a result of that page, and I consider it a success. I’d like to cook up a game that writers and anyone else can play, of burying things in their work that only the most attentive people will notice, as a way of incentivizing that kind of close attention. Then I thought, wait, that game has already been invented! It’s called life. This is also, generally, how art works. (Adam Mastroianni has a great post about art that rewards close attention.) But even if the universe already works this way, and art already works this way, we can turn up the gain, and create a subsubgame within the subgame within the game, where the rules and rewards are made more clear and immediate. Wait But Why gives one simple example: a treasure hunt with the first clue hidden. Footnotes are another.[1] For anything put in public, there is a distribution of how deeply people engage with it. For public art: ignore, glance over, stop and look, go over and look, climb/look-up-on-the-internet/copy/vandalize/destroy.[2] One of the tricky things about creating layered art is how to make it work on all levels. (Henrik thinks this isn’t necessary; I somewhat disagree.) If you put the heavy math in the main text, mathphobes stop reading. Basically, everyone is looking for an excuse to stop reading, so the more tangents and puzzles you include, the fewer people make it to the end. [3] Different people have different goals about how widely they want to appeal, and how many people they are willing to drive off, in order to engage their ideal readers. Footnotes[4], inside jokes that don’t disrupt the flow of reading, and links to explainers of unfamiliar concepts are all ways of catering to different audiences in the same piece, the same way that cafés offer both dine-in and takeout. An obvious example is text holding invisible significance. Cyphers are fun, but they tend to be fun in the same way that picking very small berries is fun. It’s fun at first, but only as a novelty, because the payoff doesn’t justify the effort. They often feel like a tangent. Can you spot the hidden tiger? (Image credit: Rusty Rust) I spent two days before I thought I found it, and felt ripped off. While we're at it, an obnoxious fortune cookie riddle: "What goes up a drain pipe down, but not down a drain pipe up?" I have been mulling this one over for more than a year. I come up with countless answers that satisfy the constraint (hot air, helium, bubbles, ivy, "pipe down!"), but none of the are the official correct answer. Many famous people make it possible, but difficult, to find their contact info 🔒. This is another example of rewarding the most engaged readers. And tech companies often put their job postings in the sourcecode, or print them to the browser console, as a way of filtering candidates. [5] Another problem with the whole concept of curiosity traps is that sometimes the people who engage most deeply are obsessed weirdos, rather than people with unusually aligned interests. This is more a problem for people who are famous personalities. In my experience, the people who write in to obscure blogs tend to be unusual in a _good_ way. Easter Eggs and gimmicks are fun, and worth including. But after trying to write an article about these artificial curiosity traps, I have realized that the best curiosity traps emerge naturally. If you write from concrete experience, and you publish a lot, people will notice the details. Because I was once a designer, every time I go to a website made by a designer, even if the layout is very plain, there’s something hard to place, but recognizably _designed_ , that I appreciate. This applies everywhere: **we notice when other people care about the things we care about.** And when we notice that someone else cares about _very similar things_[6], it’s feels like meeting a friend we haven’t met yet. * * * 1. You clicked the footnote! You are special. Now we are stuck in here together. If my footnote script supported nested footnotes, we could keep going all day! [7] ↩︎ 2. Most public art disappoints me, but the pretzeled balsa rubber band aeroplane in Stanley Park – er, I mean, _cough_ Aerodynamic Forms in Space _cough_ — has always tickled me.[8] ↩︎ 3. Case study: someone who is fairly close to my ideal reader read a draft of this essay. He pointed out design flaws with my blog, typos, and “unfinished thoughts”, and generally engaged deeply with the text[10] (even opening the devtools to inspect the footnote implementation), yet _missed every single easter egg_ , and concluded that the last paragraph was his favourite, because: > I don’t really engage with the internet as a social… as a way of meeting people. I mostly engage in other ways that are mostly fairly mercantile: to learn or disseminate knowledge, in an urgent way. This is a valuable reader who I would like to keep, even if they miss all my curiosity traps, so it’s important to make the traps subtle enough that they don’t slow down and irritate readers of this type, who aren’t interested in solving silly puzzles, and just want to get to the point, if there is one. ↩︎ 4. I mostly use footnotes like this: 1) move the weakest and most digressive parts to footnotes, then 2) realize they aren’t needed, and 3) delete the footnotes. This one got missed. ↩︎ 5. Google is famous for their billboard problem: . ↩︎ 6. A recent example: https://stevekrouse.com/readme ↩︎ 7. https://dn721608.ca.archive.org/0/items/a-night-at-the-opera-1935-HD/A Night at the Opera.mp4#t=250,285 ↩︎ 8. I am sorry that I spoiled both of these in the post about how I figured out how to deeplink :( [9] ↩︎ 9. Okay, if you're here you are very special, having figured out that I deliberately left my nested footnotes broken so you would have to view source or use reader mode. (In my markdown file, this footnote has the id #final-boss.) You have also probably the kind of person who would _definitely_ click on "Don't click this link!". Also, by now you may agree with me that puzzles and cyphers, without an adequate payoff, are an annoying tangent. ↩︎ 10. The same reader recommended the subtitle "honeypot for pooh bear". Does this not throw doubt on everything else he said? Unless, of course, it was a test to make sure I wasn't shunting the brain and blindly following all his suggestions. ↩︎

Although I’m a semi-professional musician, most days, I listen to far less music than my friends. I generally work in silence, cook in silence, run in silence. When I do listen to music, usually in the evening, I give it my full attention. Everybody is different, and different kinds of music work well in different contexts, but I can’t help feeling that anyone who doesn’t _sometimes_ listen to music the way I do is missing out. The way I do it is like this: lying on my back, wearing headphones, from an MP3 player rather than a smartphone/computer, usually in the evening in bed, after I have put down my book, but before I will be able to sleep, with, if possible, nothing on my mind. Most good music has a story. Sometimes, a song can have multiple stories, depending on what you bring to it. I like to place my mind on the melody, the the needle of a record player placed in the groove, and stick to it and see where it takes me. Melodies are stories, and I’m a sucker for stories. Song lyrics are also often stories. And a good story isn’t just the story that’s told: it’s all the other stories that have the same shape, and the story of the negative space of what isn’t said. Most stories are like fractals, and implied in the scale perceived is the resonance of the micro and macrocosm that have the same shape. But even if you aren’t interested in stories, lyrical or melodic, there are good reasons to listen this way. There’s so much to hear in a good song! There’s no way I could absorb it all in one pass through, or with it on in the background. I can listen to a song dozens or hundreds of times, and still be surprised when I really hear one part for the first time. (For more on this theme, see The Joy of Selective Attention.) I have my music collection organized into four parts: the stuff I’m tired of or never really liked (which stays in cold storage), an “inbox” of things to check out, a collection of songs that do _something_ really well that I could learn from but which aren’t really my thing, and then my active collection of songs that have turned my crank before. I usually start from the inbox (explore), and end with my main collection, or a specific playlist from my main collection (exploit). If there’s a new album from an artist I like, I try to listen to it in one sitting, without multi-tasking. I still remember the day I did that with Oliver Schroer’s Hymns and Hers, and David Francey’s The Breath Between, and in my rap phase, _The Miseducation of Lauryn Hill_ — and with Leonard Bernstein’s version of Beethoven’s 9th Symphony, which I later heard wasn’t even the best one — and was richly rewarded. I like using playlists, but depending on how the day has gone, which songs affect me varies widely. But I can usually tell which song I need to hear within a few seconds. So I find that a playlist of a few hundred songs, all of which I have moved or inspired or amused me _sometimes_ , is the right balance of being able to find the song for the moment, without spending too much time clicking fast-forward. And there’s something about (pseudo) randomness delivering me musical deliverance that I like. Sometimes, the song that would be just the thing is one that I wouldn’t have chosen, if the choice had been mine. Although I prefer to listen to songs that are good (according to me) through-and-through, there are songs that I listen to just for _one word._ I’m a connoisseur of grittily-enunicated filler “yeah”s. The rest of the lyrics might be dumb, but the sound of that careless British “yeah” in The Cars’ _Just What I Wanted_ is hard to beat. (Ziggy Marley’s _Circle of Peace_ also features an exceptional yeah.) I’ll also listen to music for the electric guitar tone (I have yet to hear an overdrive crunchier than Green Day), or the bass solo (Bakithi Kumalo on _You Can Call Me Al,_ the breakdown in _Miracles out of Nowhere_), or the snare tone, or the storytelling (Dispatch’s _The General),_ or fragments of poetry (“with his eyes he said he loved them” in _Circles Around the Sun),_ or the associations with a particular time of my life Listening to music like this is less like sitting in front of a TV, and more like a choose-your-own-adventure game. It requires sustained attention. I know it’s an active process, because if I’m really tired, to the point that my mind wanders and my heart feels sodden, then it doesn’t work. If I don’t bring anything to it, I don’t get anything from it. Such is life.

I have written before about how to link to a specific segment of a youtube video, and how to link to specific words or paragraphs within a webpage. I even knew, vaguely, that you could link to pages of PDFs. But, all in the course of a normal workday researching a blogpost, I found myself wanting reference a specific ~30 second chunk of a Marx Brothers movie from 1935, and the last page of a PDF about a piece of public art. The PDF was easy, using the #page=n syntax. As for the scene from the movie: nobody had clipped that specific part out and put it on YouTube – in fact, I didn’t even find the scene I was looking for – but I did find an mp4 on the Internet Archive with a scene that fit my requirements. It turns out, you can link to a specific segment of an Mp4, too! _This is thanks to the_ _Media Fragments URI 1.0 (basic)_ _W3C spec recommendation from 2012. According to CanIUse,_ media fragments are supported by all major browsers _(97-99% of users)._ The syntax is nice: `https://example.com/video.mp4#t=startTime,endTime` (Notice that it’s a pound/hash symbol. I always wanted to use a question mark.) An actual example: https://dn721608.ca.archive.org/0/items/a-night-at-the-opera-1935-HD/A Night at the Opera.mp4#t=250,285 What’s especially cool is that you can _**play from**_ (only specify start time): https://dn721608.ca.archive.org/0/items/a-night-at-the-opera-1935-HD/A Night at the Opera.mp4#t=250, _**…play to**_ (only specify end time): https://dn721608.ca.archive.org/0/items/a-night-at-the-opera-1935-HD/A%20Night%20at%20the%20Opera.mp4#t=,285 _…o_ r _**play a segment** (_specify both start and end time)_ :_ https://dn721608.ca.archive.org/0/items/a-night-at-the-opera-1935-HD/A Night at the Opera.mp4#t=250,285 _And,_ you can format the timestamps as seconds by using a plain integer, or as `hh:mm` or `hh:mm:ss` . (You can also, weirdly, use `h:mm:ss` format.) This is generally more convenient, since most video players reference current playhead location using hours and minutes, not seconds. https://www.autodidacts.io/content/media/2026/09/2026-09-24-copy-all-tabs-urls-firefox.webm#t=00:00:04,00:00:05 _And_ it works for audio files too?! Look at how much good stuff I have been missing out on! While we are at it: you can also crop SVGs. I didn’t know that, either. At this point I’m just duplicating the excellent, slightly more intimidating article on MDN, so if you like what you’ve heard, go read it. Even better, try the playgrounds: link, link!

Once upon a time, I ran across an experimental project called TabFS, which mounts browser tabs as a filesystem. This seemed like such a cool idea! Ever since, I have wanted a way to interact with tabs programmatically. But the main thing I’ve wanted to do is be able to grab the URLs of some or all of my open tabs, and paste them into a text file. Each time I have looked into this before, it hasn’t been possible without installing an extension. And I don’t _like_ installing extensions. I often have text files of URLs, one per line, related to some project. Often I want to be able to open them all in Firefox (easy in Helix editor with `%s\n<ret>xgf` [1]) while I’m working on the project, close ones that aren’t relevant any longer and follow new leads, and then save the updated list when I’m done, without interacting with bookmarks (my graveyard for URLs that are “interesting” and “relevant” and I’m definitely never going to visit again). Another common use-case is when I want to copy the URLs for a group of tabs and pass them to a script (in this case, my Ghost API markdown export script). Today, I discovered that it’s not just possible, it’s easy! It looks like this feature was added in **v152.0a1** , 4 months ago. 1. Select some tabs (right click on a tab, then click “Select all tabs”; or click on one tab, then shift click on another to select range, or ctrl click to cherry-pick individual tabs) 2. Right click on tab, or open File menu 3. Click “Share” 4. Click “Copy _n_ links” Watch: 0:17 /0:19 1× **You can do steps 2-4 in one go by pressing (in sequence) Alt f h l** I want to be able to do _everything_ with the keyboard. However, although you can cycle through tabs and move tabs with keybindings, weirdly, there doesn’t seem to be a Firefox keybinding for _selecting_ tabs, or for opening the tab right click menu (Shift+F10 opens the page right click menu, even if tabs are selected and the mouse is hovering over a tab). If you paste these tabs into a rich text field, they will be links with titles, but if you paste them into a plaintext editor, they will be plain URLs, one per line! * * * 1. `%` select buffer, `s` open split dialog, `\n` split delimiter, <ret>/Enter split selection on delimiter, `x` extend selections to whole line, `gf` (g)oto (f)ile, which opens URLs in the browser ↩︎

_[ In_ Iteration_, I wrote about the mistake of demanding_ painting lessons from van Gogh_. This essay tugs upon on the same thread. ]_ There's an old saying that in order to teach a dog, you just have to be smarter than the dog. As with most proverbs, I think there's some truth there, but it's not that simple. 1. If the path is straight, to lead the dog, you just have to be ahead of the dog. 2. If the path is non-linear, mentorship from someone just ahead of you might help, but also might lead you on unnecessary detours. As far as I can tell, **approximately nobody's development follows a straight line toward a destination**. Even with simple skills, like, say, swimming a certain distance faster, following the advice of someone better than you might have suboptimal results. Because it's possible to "improve" using naïve optimization that makes you better in the short run, but worse in the long run. A slightly more advanced student might try to improve the floundering student's back crawl, since that's what they're using, poorly, and the coach is using more skillfully. But anyone with actual experience will try to figure out why the student can't breathe properly doing front crawl. First it will be slow, then it will be fast, which, in the long run, is much better than getting quick gains and then plateauing at a local optimum (very fast at a very slow stroke). _(A wiser mentor, even if they took a circuitous path, might, in hindsight, be able to see that the path they took isn't the best path. However, it's also possible that the shortest/fastest path is not the best one for a particular student, for reasons that are opaque even to the wise mentor. Perhaps, if the wise mentor were even wiser, they would see that the detours are, in this case, necessary. So, in the case of my illustration, a wiser wise mentor could have taken the direct route_ or the spiral route _, and might recommend the route they took_ or the other route _, depending on circumstances. However,_ as long as the game is in play, there's perhaps a limit to the knowledge of even the wisest of the wise_.)_ But getting a fancier coach isn't the whole answer, because of the problems I mentioned in Iteration: the scarcity of expert-time, expert-access, and expert-patience, as well as difficulty stepping down expert advice to a level that can be understood and applied by the student. (The difference between an expert and a teacher is, I suppose, whether the stepping-down is bundled with the expertise.) My solution is, go ahead and learn from the person just ahead of you, but verify what you've learned with a true expert. When I did my lifeguarding pre-requisite course, I found coaching from a more advanced student more helpful than coaching from the instructors. The advanced student had _just learned_ what I was trying to learn. It was fresh, rather than muscle memory. The instructors would explain some new stroke: I’d try it, and it would totally fail to work. I'd watch the instructors do it, but they made it look so effortless I couldn't even tell what was going on. So I'd ask the more advanced student if he'd gotten it to work, and he'd say, yeah, by ignoring what the instructors said, and doing what felt intuitive. I managed to get the imperfect intuitive version to work, which got me from zero propulsion, to enough propulsion I could continue practicing the stroke without sinking. If I continued to just do that version, I might end up with some bad habits. But from there, I could try to make it look more like what the instructors were doing, and then ask them, _does this look right?_ * * * In the AI world, the cool kids often run cheap dumb models to do the grunt work, and smart expensive models to do the planning and the evaluation. In business, it's the same, with high paid people monitoring and correcting, and low paid people doing the work. (As much as it pains me to suggest that the AI and business junkies are onto something, they are. But it's an idea they stole from the structure of the human brain and/or an ant colony, so we don't need to give them credit.) We need multiple mentors, some far ahead of us, and some just ahead of us. That way, I can spend most of my time learning the practical details of the skill from the person just ahead of me, and then occasionally check in with the real expert, to make sure I'm going the right direction, and haven't ended up copying the intermediate student's flaws along with their skills.

The first thing I wrote when I started this essay was, "Written 2026-09-01." Then I wrote "Idea long before that". Then I opened my post ideas file and searched for "metadata", and found the line where I had written down the title for this essay, four years ago. Then I wrote "Idea and title 2022-07-30". Does this sound tedious? It is tedious. But, it's worth it. I have written before about how much I like dating. (No, no, not _that_ kind of dating. ISO_8601 dating!) But my obsession with timestamps is just part of a broader belief in the long-term value of metadata. Why save metadata? Briefly: _because it's**way** easier to record data when it's fresh, than to have to piece it together later._ Additionally, I think that: 1. We wildly overestimate our chance of remembering things. 2. We often have no idea what will prove to be important. We think, oh, that doesn't matter. Or, oh, I'll never forget _that._ But it does, and I will, and I'll only realize that it does after I have. More likely, someone _else_ will realize it does, long after I'm gone. And they'll be like me: wasting a non-profit's money and getting a headache trying to trace the provenance of obscure foreign-language manuscripts that were published under different titles in different languages. What does this mean in practice? Here are some examples: * * * I write songs. When I write a song, I make sure to: 1. record a memo of it 2. write down the lyrics 3. put the chords in the lyrics 4. fill out a template with all kinds of boring stuff about it: the key, capo position, accompaniment, date started, date worked on, date finished, tempo. Just looking at it makes me tired At the time the song is fresh, it all seems a bit absurd. Am I really going to forget _what instrument I play_ on a song that I wrote? A song that took me, probably, a day to six months to finish? Am I going to forget what key it's in? Forget what order the three chords go in? Yes, yes I am. And years later, the drummer is going to forget he played drums, and try to play cello on it, and it will sound bad. Some songwriters I know don't put the chords in their songs, or notate the key. They are the creative types that write in a cute little notebook, and haven't written custom software for managing their song catalogue. Then, years later, they try to play them, and discover that they have forgotten which order the three chords go in. (Fun fact of interest of to almost nobody, because the Venn diagrams of "songwriter" and "nerd" have so little overlap: Regex can make adding chord notation to lyrics way faster.) Also, lyrics will drift. Usually this is good. Unmemorable parts get dropped, just like with folk songs. But sometimes you want to go back to the first version, in case something good got left by the wayside. * * * Code comments are another obvious form of metadata: this is from actual code, sadlywatching a neanderthal rub sticks together trying to make fire I write a lot of infrequently used shell scripts. Some of them are meant to be run from a particular place, or as a step in a pipeline, or are full of known footguns that I haven't bothered to fix. It saves a lot of time to add a comment at the top with usage instructions, while I still remember how to run the script. Better yet, _echo_ the usage instructions, and exit if the config is wrong. I also think of logging as a form of metadata. As soon as something breaks, the first thing I do is add logging. So I've started by adding logging right at the beginning, so all I need to do when the software breaks is turn on the logging. If you have gone to the dark side, I highly recommend that you record: * what model you used * if running locally: quantization, inference engine, and parameters * the prompt The thing about vibecoding is that I don't understand it, so I don't remember it. This paper trail allows me to remember where I was on a project, if it gets shelved and then I try to come back to it. * * * If you're a writer, have a good schema for tracking different versions of things. Fancy people use git; I use numbers at the end of the filename. `Title final-final-FINAL.doc` does not count as "a good schema for tracking different versions of things". Any time I write an email in a text editor, and then paste it into my email client before sending, and then tweak it in the email editor, I add a note to the text file NOT UP TO DATE. (Sometimes, I put CANONICAL VERSION in whatever _is_ the up-to-date version of a document that lives multiple places, in different forms. This is, of course, fragile, but it's often better than nothing.) When I scan papers, I pencil on them that they have been scanned (or, typed in), and the date. * * * Another reason to save metadata: to protect against lawsuits. Often, I write something, or write a song, and then notice a similarity to some other work. If I saved a good record of my process, I can cross-reference: had I read that article, or heard that song, before I started mine? (Astute readers will notice that this won't actually protect me from lawsuits, though maybe it would help. But it will ease my conscience.) Obviously, if you are into plagiarism, then you probably _shouldn't_ save metadata, since it will make it easier for them to catch you! * * * Another kind of metadata is **indexing metadata**. It's not data about what, it's data about where: lightweight, easy-to-find data that points to heavy buried (meta)data somewhere else. Things like unreasonably long filenames, and notes about where the notes are. * * * What are the reasons not to save metadata? * It's slow * It takes up space * It's tedious * It's revealing These are legitimate reasons, but in most cases, I think they are outweighed by the benefits. **Metadata adds the fourth dimension to your projects: their trajectory through time.** It's much easier to steer something into the future if you know the path it took on its way to the present.

I like building things: it makes me happy. Learning and discovering things is a close second, but building is more satisfactory. Since I’m not that good at actually, you know, _building_ things, the thrill of “building”, in my case, has mostly come from writing software. Software is one of the easiest ways to create something satisfyingly useful out of nothing. For a long time I’ve known, at some level, how much joy building gives me. [1] But it really sunk in while working on two recent hobby software projects. They were both dumb little Python scrapers that I cobbled together, that did something useful that I’d wanted for a long time. (Link, link.) They were pretty basic, but gave me so much irrational pleasure. It was the most fun I’d had programming in years. [2] Then, _[cue Jaws theme song]_ AI came along. Like pretty much everyone else, I’ve been cautiously experimenting with AI-assisted coding. We are past the point where we can easily dismiss AI as “stupid” (though it is), or “doesn’t actually make me more productive”. So, in classic motte and bailey fashion, I am retreating, while pretending not to: AI coding is _less fun._ This seems counter-intuitive, because vibecoding is _absolutely thrilling …_ AT FIRST. Refactoring vibecoded software? Not thrilling. I’m not going to claim, as some do, that vibecoding doesn’t give the thrill of actually building something. Because it does, partly. I think that one of the things that’s driving the adoption of vibecoding is that it **frontloads the fun** , at both the project- and career level. The Vibecoding Thrill Curve If I have an idea, I can usually get a working prototype _very_ quickly, with very little effort. Instant dopamine! Similarly, someone with no coding experience can “code” an app, and it will probably work. Instant dopamine! I have a personal policy against frontloading the fun. So, I don’t like this, just like I don’t like getting paid for work I haven’t done, going into debt, and arrangements where someone else does my work today in trade for me doing their work tomorrow. I posit that there are several distinct sources of thrill/satisfaction when building something: 1. The thrill of the idea 2. The thrill of making the idea real 3. The satisfaction of hard-won accomplishment 4. The satisfaction of a job well done 5. The thrill of learning 6. The satisfaction making something that solves a problem you have Vibecoding, in general, can give you 1, 2, and 6. But 3, 4, and 5 are not natural consequence of vibecoding. In tenuous defense of vibecoding: I can, and have, built things with AI that I either never could have, or never would have, built otherwise. Mostly, it’s cases where it would not be worth me spending weeks building it by hand, but if I can build it in minutes or days, it is worth it. * I one-shotted a _has-this-already-been-submitted-to-hackernews_ bookmarklet, based on Taylor Troesh’s Discuss on HN snippet. It gave me significant satisfaction, solved a problem I had, I use it regularly on my Wallabag favourites page. * In very little time, I vibecoded “Party Market”, a prediction market simulator party game to go along with a flash lecture on prediction markets, and ran a market on “how long will I drone on for”, live, during my lecture. It turned a dry and boring lecture into grade-A entertainment. The code was garbage and it was full of bugs, but I never could have done it writing it by hand in that amount of time. It wasn’t a choice between doing it with AI and doing it by hand: it was a choice between doing it with AI _and not doing it at all._ * When a fire chief I know was going to invest way too much money in an “AI appliance” that didn’t even say which Nvidia card it had, or what model/generation of Xeon it was, I vibecoded _with free plan Lumo_ , in two afternoon sessions, a functional Pyronear based Wildfire Early Warning System that ran in realtime on a live feed on my laptop, and was able to detect the first frame of smoke on the ignition footage I fed it. * I built a browser-based Kobo database migration assistant (link), so that people unfamiliar with SQLite can move their books and annotations to a new device. * When a singing lesson with my vocal coach fell through, I built the realtime pitch training app I’ve always wanted — and have planned to build for about a decade. In most of these cases, I was immensely pleased with the results, but the process wasn’t _fun_ the way writing my dumb little scrapers was fun. Of course, handwritten software development isn’t always fun. My first Golang project was, for large sections of time, with a deadline, certified Not Fun™. Fighting the Rust compiler the first time I tried to write, like, twenty lines of Rust code was Not Fun™. But it was, in the long run, Type II fun, because it gave me the sense of accomplishment, and the glorious _I did it!_ feeling (And, next time, it would be easier.) I’ve tried everything from local models running in llama-server, to copy-and-pasting from free, no sign-up services (Phind, Lumo, etc) for one-shot scripts, to full on remote server Clod Masterpiece 200™ all-you-can-eat plan with custom-written auto-approval hooks that were worryingly close to `–dangerously-skip-permissions`, developing algorithmic trading models. **It’s all the same** : it’s “productive”, but I don’t like the way it makes me feel, and I don’t like the quality of the output. It manages to make me feel dumb and superfluous (and, sometimes, dirty for even using it), without earning my respect. The word “dopamine” made me think: vibecoding is a bit like a series of first dates or one night stands. You don’t develop a relationship with the project: you don’t get the same kind of warm fuzzies, oxytocin, attachment. The problem with a dopamine-only life is that thrills become dull with repetition. Another way to look at it: vibecoding introduces a layer between you and reality. I am the kind of person that thinks, in general, reality is _more fun_ when experienced first-hand. Vibecoding is like being an executive of a large company that is barely under your control. You never feel like you _really_ know what’s going on, and there’s always the suspicion of sabotage and loafing. Things appear fully formed, but not quite how you imagined. Sometimes they work, and sometimes they don’t, and there’s nothing you can do about it, other than pouring more money into the black box, and yelling louder. So that’s what you do. _This essay was outlined in blue ink by candlelight. AI wasn’t even allowed to correct my typos: savour them._ * * * 1. As a kid, I wanted to be an inventor, which neatly combines building and discovery. Writing and songwriting, my other main activities, are also forms of building that, like software, require mercifully little contact with physical reality and other humans. ↩︎ 2. I’d mostly been writing shell scripts, and Bash is not known for being _fun_. ↩︎

A lot of thought and talk goes into how good things are, when they’re new, and the review industry is big business. A smaller amount of thought goes into durability, but some people do think about it: there are products with lifetime warranties, the buy it for life movement, and human longevity research. Longevity researchers have discovered that “lifespan” is a misleading metric. Often, people die gradually, not all at once. Sometimes, people can’t even decide whether someone is dead or alive: it’s that ambiguous. So a single number doesn’t really cover the decay curve. “Healthspan”, and disability-adjusted life-years, became the new metrics. (I think we could go further, and coin some kind of meaning-span — the time someone lives before becoming jaded and giving up — in order to codify the Jack Londen-esq _I’d rather be ashes than dust_ sentiment.) But in the case of physical things, we are still stuck with concepts that map more closely to “lifespan” than “healthspan” (let alone meaningspan). But the _health_ span of products is as relevant as it is for humans. Different materials decay in different ways. It’s an oversimplification, but, broadly, cotton fails beautifully. It gets _softer_ , and denim, according to the people who buy jeans that have been destroyed before they even got them, falls apart in alluring ways. It fails in a fairly predictable, linear fashion. Same with copper and brass and silver, which develop patinas that some people like, and wood and leather, which weather fairly well. (Wood does rot, so that’s an exception.) Manufactured materials, in contrast, tend to get ugly with age. Polyester pills up, and then bursts. It shrinks in the dryer until pants cling like shrink-wrap. Spandex bags out grotesquely, while the other fibers try to hold the garment together. Plastic patio furniture develops that cell-y sunburned I’m-turning-to-dust layer that is ick, while wood just turns grey and slimy like decaying materials are meant to. But material is only part of it. Quality also affects decay patterns. A well-built thing will tend to decay more elegantly than a poorly built thing. A poorly built thing tends to have weak points that fail dramatically. Well-built things tend to just wear down. Immanuel touched on this in his article on joyful frugality: sometimes, being cheap is expensive. But when you buy a pair of cheap polyester pants from Costco, the tag doesn’t say, “looks great now but will be a pill festival in two weeks!”, or “$3 per wear and then a lot of stress about it not being recyclable”, or “2-for-1 jeans that turn into tights in the dryer!!!”, or “advanced self-disassembling model”. Software is like this, too. Vibecoded apps that are packed with features tend to explode into a mosaic of loading icons and error messages when you try to actually use them. BSD, from my limited experience, is like a slower, jankier version of Linux, but the way it fails is _so much more fun_. It just slows down. You can watch the re-paints happening in real time. It’s a very small motor with a very large gearbox, that just keeps going uphill. Linux has more motor, but less gearbox, so it goes fast until smoke comes out of the hood, but then it doesn’t go at all. Modularity also matters. If something can be replaced piecemeal, it’s more likely it can be repaired. If electronics fails, and it’s surface mount and potted in some nasty goo, there isn’t all that much hope (I’ve never managed to fix a quartz watch). But mechanical watches, and boots that can be re-soled, and laptops like golden era Thinkpads and Frameworks, with multiple different modules that plug into each other, can, though repeated Phoenix-esq rebirths, become objects-of-Theseus. And, yes, humans are like this, too. There are NetBSD people: not fast or glamorous, but just as functional when under extreme load as any other time. And then there are polyester people, who look great in the store and fall apart two weeks later. When shopping, building, and dating, consider: will I still love this when it falls apart? Because that’s a larger portion of the lifespan of products and people than it may seem.

Some moments are intensely vivid and pleasant. Other moments are intensely and vividly unpleasant, or, more commonly, unmemorable and weakly depressing for no reason. The types of experience can be separated, on lucky occasions, by a matter of seconds, and minimal change in external stimuli, which reveals the fraud or one or the other if not both. While meditating for an ambitious seven minutes, I had a thought about this so interesting that it was kind of the end of my meditation. It was, _this is why boredom is good_. (I had, in fact, been bored for hours before I started meditating: the seven minute sit was just the cherry on top.) My theory is, boredom works like this: first, the mind rushes around stirring up dust and fetching pieces of flotsam and pointing at things and worrying, trying to find or generate something “interesting” despite the lack of novelty. This can go on for quite some time. **But then it gives up.** If the mind was a dog, it has lain down and rolled on its back with its tongue out, sleeping or dead. This is when the Just Being™ starts. (I know, I’m like two thousand years late to the party; and also, I’m already blogging about meditation after a seven minute sit with my belly full of whipped cream and chia pudding. Please ignore my lack of credentials; something my long-term readers have become skilled at.) The thing about Just Being is that it is inwardly and outwardly pretty much like being bored, _except it’s not boring_. For whatever reason, top-quality Just Being feels more important, more intrinsically valuable, even more _urgent_ that just (notice the lack of capitals) doing. Boredom tends to produce daydreamy “good ideas” as the mind unspools, recombining recent impressions in potentially novel ways. When I go camping for a few days with no internet, I find that I am sometimes hot and bored and sweaty at first, and feel like I’m not “accomplishing” anything, but by the end, my whole life is reënchanted, and I return to civilization filled with ideas for new projects. But I think the value of real boredom, as well as the best ideas, the ideas that are from a completely different frame — I should quit my job kinds of ideas, rather than here’s a way to do better at work kind of ideas — has less to do with the decaying pseudohypnagogic associative activity of a mind driven to distraction with boredom, and more to do with this sudden contact with reality when the mind gives up on being entertained. This space intentionally left blank:

Some people’s clothes have a way of riding up in unfortunate places, in a way that seems almost willful. It’s as if the clothes are sentient, and cloth and wearer are locked in a lifelong battle for supremacy. I know people who can look out-of-place in a tailored jacket. In fact, I am one of them. I know other people who are the opposite: they can get away with anything. Cloth, no matter how cheap or poorly cut, drapes over their muscles. If I am exhibit A, Immanuel is exhibit B. He can look stylish wearing things that I’d be ashamed to even be seen in public wearing. He had a pair of boots that he’d repaired with bicycle inner tubes and brass rivets. The repaired section immediately blew out, but he wore them all over town, to special events, as if they were the latest hip thing, and because he pretended, and maybe even believed, that they _were_ the latest hip thing, they were, and everybody commented approvingly. This is, of course, totally unfair. And, since, in this case, I am on the wrong end of the bargain, I want to know _how it works_. Hypothesis time: **One** : some people are just the wrong shape. In so far as clothes are made for an average-size average-shape person, some of us are just plain unlucky, by being non-standard. We rattle around in our clothing like a metric allen key in the head of an imperial allen screw. I have long arms, and large ~~man-boobs~~ _pecs_ , and wide shoulders, and a super huge neck. There is not a single off-the-rack dress shirt that fits me, and I’ve looked. This is a nice theory, but I think it’s at best partly true. **Two** : suave people are always adjusting their clothes without us noticing. Maybe everyone’s clothes, without maintenance, look equally ridiculous, but, somehow, suave people are more aware of how ridiculous they look, without seeming to be preoccupied with their clothes, or having it take them out of the moment, and are thus able to be constantly making image-maintaining micro-corrections to their outfit. The difference between the person with the wedgie and the person without is that the person without pulled their clothes out of the wedgie before anyone noticed. I think this is probably partly true. I’ve noticed that glamorous people are often continuously touching their hair, in a way that is totally pointless, other than as a way of attracting attention to their hair, and as a result, attracting a mate. So, if your clothes are nice, and you look like, constantly adjusting them could have a double benefit. **Three** : goofballs aren’t embodied The people with clothes that never seem right are often the people who are clumsy, or absent-minded. The people who always have the just right outfit are often more _embodied_ in other ways (like, actually going the gym). Is this a coincidence? Maybe it’s simply that some of our minds don’t tend to notice the sensations from our bodies, because we are (obviously) super busy thinking about important things (like why other people look better in their clothes than we do), so we don’t notice the tactile sensation of fabric bunching up and stretching, or don’t connect those sensations to a need to do something about them. **Four** : it’s a psyop Maybe there is absolutely nothing physically different about the appearance of the suaves and the suave-nots. It’s all in how we perceive people (and likely, how confident they are, which affects how we perceive them, which affects how confident they feel… unfortunately). In this case, Immanuel can get away with his hideous boots, because everybody is lost in his eyes and full of warm fuzzies and social bonding neurotransmitters. But when they’re talking to _me_ , they’re bored and uncomfortable, so they look at my feet, and think, _dang!_ that guy needs a new pair of boots! I think it’s definitely partly a psyop, and we shouldn’t let them get away with it.

Consider the dishwasher. Dishwashers are great: I really, really like having a dishwasher. Growing up, we didn’t have a dishwasher. Much time was spent immersing plates in soapy water. Now, a house that doesn’t have a dishwasher seems _lacking_. I recently moved to a house that lacked a dishwasher (now, there's a small mobile unit that I complain about constantly). Much as I disliked not having a dishwasher, adding a dishwasher hardly saved any time, because, following the timeless freemarket principles of supply and demand, dish usage soon caught up with dishwashing capability. The equilibrium — time spent immersing objects in soapy water — was maintained, even as thruput increased. Rather than saving dinner dishes for dessert, all kinds of culinary extravagances cropped up, like drinking wine out of wineglasses instead of mugs, and eating small objects on fussy little saucers, and so forth: extravagances which anyone who has gone camping, or lived without a dishwasher, will know are optional make-work. (Colanders are still outré, thankfully.) This, in case you didn’t notice, is a metaphor for AI. Because these days everything’s a metaphor for AI, and vice versa. We used to write all our bugs by hand, and then fix them. It was very slow. Now, AI writes our bugs for us, and fixes them. The amount of code generated is, indeed, quite large. However, time spent working has not seemed to decrease. Shouldn't we all be sitting on the beach drinking out of coconuts now, like we’re designed to do? Hasn’t that been the dream since the industrial revolution? And hasn’t it always remained just that: a dream? I don’t know whether I like efficiency because I like programming, or like programming because I like efficiency. In any case, optimizing optimization is endemic among engineers. The idea is, you automate the boring stuff, and then get to work on the important/interesting/fun stuff (like babysitting Claude while it has spitups). But I can’t help noticing that the faster we speedrun optimization, the faster the treadmill goes, and the whole system begins to seem zero-sum. This, I think, is a key insight that the Amish beat most people to by several hundred years. Optimization is great for “productivity”, whatever that means, but seems orthogonal to the quest for eudaemonia. In case you haven’t noticed, despite being what is officially known as a lazy bum, I am a card-carrying member of the school of Do Hard Things: the meaning over pleasure camp. Because, as far as I can tell, pleasure alone is hollow, and, also, relative. Paradoxically, suffering can make joy more intense. Hence the ultramarathoning and cold water plunges. Given that I believe that Hard Stuff (at the right dose) often makes us tougher in the long run, and resilience makes us happier than mere ease, here’s a question that nags me almost continually: **what hard stuff should we optimize away?** The normal answer is, _all of it!_. But I’m not so sure. I’ve met survivalists who almost starved because it took them so long to hunt a bear; homesteaders who make their own cheese, and spend most of their time meeting their own needs for food and shelter, with minimal interchange with the outside world; and executives who don’t even clean their own house or write their own emails. And, frankly, it doesn’t seem like the executive lifestyle is maxxing out Maslow’s Hierarchy of Needs as hard as other lifestyles. In what sense is it optimization to pay someone else to chop your firewood, and then turn around and pay the gym for weights and a room to lift them in? But this is just part of the question. If there are strenuous activities that are enjoyable (like chopping and stacking wood, in moderation), then it’s pretty obvious that just doing them makes more sense than delegating them and then needing to get this bourgeois thing called “exercise”. But what about psychological hardship? The Stoics are big on making your goat ungettable. And, as all skydivers know, practice makes perfect. So, obviously, the way get an ungettable goat is to surround yourself with an army of goat-getters and fight the good fight ceaselessly until thy goat is ungettable, amirite? So (taking the argument to its conclusion), the too-small dishwasher with the pump that makes chewing-on-broken-glass sounds, and the obnoxious friend, and the horrible politicians, and the sun with no dimmer and the earth with no thermostat are all part of the grand journey to un-goat-gettable Stoic perfection. Or, is it possible that if I _fixed_ the things that bother me, the universe would graciously provide even more skillful goatgetters for my personal benefit? Or, at least, that I would be able to be annoyed by new, bigger things? Perhaps, just has perfectionism is (often) optimizing at the wrong scale, holistic optimization needs to account for the fact that sometimes bugs have a role in the ecosystem. Between the Scylla of overoptimization, and the Charybdis of waste … there’s a whirlpool of Scylla and Charybdis metaphors, which we just got sucked into.

I have been informed by the internet that using Ollama is passé. Though it’s easy to get set up and use, and popular with rank beginners like The Autodidacts, power users balk at the fact that: 1. Performance is worse than llama.cpp _et al._ 2. It’s a VC-backed startup that’s rapidly selling-out on the local-first promise by touting paid cloud offerings 3. It relies on Llama.cpp for the heavy lifting, without giving credit where credit is due Long ago, I thought, okay, I should just switch to llama.cpp. I’ve used whisper.cpp, how hard can it be? Let me just say: there’s a _reason_ Ollama is so popular. Every time I tried to switch to llama.cpp, it either a) it wouldn’t compile, or b) I got cryptic CUDA memory allocation errors. (Even when I went to run the commands to test for this article, it was broken, because of an interrupted brew upgrade.) But now I am past that. After fruitless Googling, I threw command line arguments at it until it worked, and now this post is fruit for you to Google. ### Step 0: Pre-requisites You will need a system that can run the model you want to run. Two websites that are useful for getting a general idea of what’s likely to fit (though they aren’t the last word!): * https://www.canirun.ai/ * https://willitrunai.com/ (Which one is ripping off which? I can’t tell.) For this post, I’m assuming 16gb of VRAM. You will also want a fast-enough processor and plenty of system RAM, since we will be offloading some of the work to CPU + system ram, and some to GPU + VRAM. (Find out how much VRAM you have with `nvidia-smi`, `rocm-smi`, or `lspci -v | grep -i vga -A 12`) ### Step 1: Install Llama.cpp {#step-1-install-llamacpp} Once you figure it out, compiling it is easy. But llama.cpp changes so fast, I prefer to use the version packaged by brew, for simplicity and automatic updates. Once you have Linuxbrew installed: brew install llama.cpp As long as your $PATH is set correctly, llama-cli and llama-server will now be available everywhere. ### Step 2: Download the Model Obviously, we’re going to be using a quantized model. 4 bit quantization seems to generally be considered a good compromise, and I’m not fancy, so I went with the crowd. Llama.cpp can download directly from HuggingFace. Unsloth seems like a reasonably trustworthy provider, so I went with unsloth_Qwen3.6-35B-A3B-GGUF_Qwen3.6-35B-A3B-UD-Q4_K_XL.gguf and mmproj-BF16.gguf. (FWIW I have no idea what I’m doing. You should probably go read Simon Willison or something. Except, they don’t write about stuff this basic. So you’re stuck with me!) ### Step 3: Make it fit! After reading various things about offloading the MoE (mixture of expert) layers to CPU, and trying all kinds of things that didn’t work, I found one that did: offloading only ~16-24 of 40 layers to GPU. Otherwise, I got CUDA allocation errors (ie, not enough VRAM). The relevant flag: --n-gpu-layers 16 # or -ngl 16 I was able to get up to 29 layers on GPU. 32 was too many. You can calculate exactly how many you can get, if you want (layer_size = model_size / num_layers), but it also depends on what else is using VRAM, so aim conservative. It’s also instructive to see what Llama can find in the way of devices. It might be `CUDA0` or `Vulkan1` (or possibly `Klingon2`). Find out what you’ve got with: `llama-cli –list-devices` ### Step 4: Run it! Here’s the full command (adjust the paths to match where you downloaded the model): llama-cli --model /path/to/models/Qwen3.6-35B-A3B-GGUF/unsloth_Qwen3.6-35B-A3B-GGUF_Qwen3.6-35B-A3B-UD-Q4_K_XL.gguf --mmproj /path/to/Qwen3.6-35B-A3B-GGUF/mmproj-BF16.gguf --n-gpu-layers 24 -p "Write a 100 word sermon on the moral and practical benefits of supporting independent bloggers such as Autodidacts.io with micro-donations, as a thank you when they write an article that improves your life in some tiny way." Here is the turgid homily Qwen produced: _Beloved, heed the call of gratitude. When an independent voice on Autodidacts.io sharpens your mind or eases your burden, however slightly, return the blessing. A micro-donation is no small matter; it is moral alchemy, transforming gratitude into justice for the creator._ _Practically, these scattered offerings fuel the engine of truth. They ensure the lights remain on, servers hum, and wisdom flows free from corporate chains. By tipping for tiny improvements, you sustain the ecosystem of ideas. Support the independent blogger. Honor the craft. Let your thanks keep the wellspring fresh. Amen._ It’s so bad, it kind of makes it _better_ , because it makes its own point! ### Or run a jazzy web interface It turns out that `llama-server` is super easy to run, and much more pleasant to use. Really, the only reason I can think of to use llama-cli is for one-off questions and scripting in bash pipelines, when you don’t want to write a Python wrapper. Start it like this: llama-server --model /path/to/models/Qwen3.6-35B-A3B-GGUF/unsloth_Qwen3.6-35B-A3B-GGUF_Qwen3.6-35B-A3B-UD-Q4_K_XL.gguf --mmproj /path/to/Qwen3.6-35B-A3B-GGUF/mmproj-BF16.gguf --n-gpu-layers 24 Then, go to 0.0.0.0:8080 (the default host and port), and you get a nice web UI. I didn’t even know it existed, but it does, and it’s as good as Jan.ai or Lumo. ### Conclusion I’m getting 14-22 t/s, which is pretty feeble, but adequate for my limited needs (mostly, looking up syntax I’ve forgotten when I’m offline). This is probably because my CUDA install is, as usual, broken, and I’m using Vulkan. _[Every time I buy a laptop,_ I decide that next time I’m buying AMD graphics_. And then, because I’m a cheapskate, I buy Nvidia, and spend the next half-decade fighting with my graphics drivers.]_ Soon, I’ll write about using Llama.cpp for handwriting OCR. (Spoiler: Qwen 3.6-35B-A3B works even better than Qwen3-VL:8b.) Other people use the 8 bit quant of Qwen 3.6 27B, and like it. I don’t know where the sweet spot is between a bigger model with more aggressive quantization, and a smaller model with less aggressive quantization. I haven’t tried 27B Q8 yet, but I probably will soon. If you’ve tried both, let me know your impressions!

Every one’s talking about their agents. What agents are we talking about — secret agents? casting agents? foreign agents? literary agents? cleaning agents? _(this slop is just_ full _of ionic and non-ironic surfactants—)_. Oh, oh, _AI_ agents. I dislike ambiguity except when I don’t. AI agents is clumsy, and begs to be shortened. Agents is the obvious contraction, but when you drop the AI, it becomes annoyingly ambiguous out of context. It is a canker that eats previous usages of a useful word! So, let’s call them aigents. Let’s call it, _aigentic engineering._ I’m surprised no one has coined this portmanteau yet; or, if if they did, it never caught on. It’s totally unambiguous, and it’s a fresh namespace. Existing secret-foreign-travel agents needn’t get mixed up in this. It’s weird, but we’ll get used to it. And when we get bored, we can shorten it to “aigs”.

Sometimes people say goodnight by saying, _see you in the morning,_ if that’s expected. This bugs me.[1] While I certainly _hope_ to see them in the morning, and consider it highly probable, even in a modern world, it isn’t guaranteed. To me, all goodbyes are an acknowledgement of this fact. They exist, mainly, for the unlikely case that this _is_ the last time. The intensity and duration of the goodbye reflects the value of the relationship, times the probability that it’s the last goodbye. Thinking like this isn’t about fixation with death, it’s about appreciation of life. In my goat-like focus on whatever I’m working on, it’s easy to ignore the transient beauty all around me. Awareness of mortality — my own and others’ — is the antidote. When you have a rose bush in your yard, or a vase of cut flowers, you don’t think, _oh, that’s nice, but I’m in a hurry; I’ll stop and appreciate that tomorrow._ (Or maybe you do: but that’s lame.) Because it might not _be_ there tomorrow. People often seem more durable than they are. In any town, there are reminders of this, and somehow it still seems like a surprise. Grandma; the obituary for a 40-year-old single mother, with no cause of death; the fiddler who gets in a car crash a week after the jam. There are also people that just … _drop_. On the stop signs and people’s laptops in my community are stickers, “What would Lily do?”, after a beloved community member — a vibrant young adult with no preexisting health conditions — collapsed playing volleyball, and never got up again, leaving all the friends she used to bring together, and two grieving dads. There’s the old guy who ran the second hand shop: a dirty old man in the most literal sense, a fountain of information and opinions, ensconced in a pile of junk and papers and half-eaten toast buzzing with flies. Even the kind of person who seems indestructible, like a piece of the town itself, is gone one morning, the shop boarded up. I was in the city for a a few days. In the lobby, a couple stood, amused, watching as we took cheeky photos in formal wear. The guy, twinkling, said it looked like an album cover. We said we actually were a band, and got talking, and then went our separate ways. A few hours later, at a random intersection of the city, we meet the couple again. They are about to get married: they are staying right across the hall. (A man bent double from fentanyl poisoning, holding a Tropicana orange juice bottle that probably contains something else, shuffles by on a skateboard: the contrast with the mood of the moment is poignant.) The guy were are talking to is one of the happiest, kindest, twinkliest-seeming fellows I’ve met. I ask if he’s a meditator: he is, a fairly serious Tibetan Buddhist. I ask him what’s the secret to his happiness. We stand in the rain for probably half an hour, talking. Two weeks later, the woman comes to a concert in a different city. The man is dead: an accident. But she doesn’t know if it was actually an accident. It may have been suicide, which was how he had just lost one of his best friends. I only knew him for about an hour, but I still think about him often. There are others in the same category, still with us. But I only every learned their first name, and the world is a big place. * * * Two of the first songs to move me deeply, when I was a child, both spoke of impermanence. One was _Big Yellow Taxi,_ by Joni Mitchel. Specifically, these lines: > _Don’t it always seem to go > that you don’t know what you’ve got, ’til it’s gone > they paved paradise and put up a parking lot_ [2] Another was _On The Road Again,_ by Willie Nelson: > _On the road again > Goin’ places that I’ve never been > seeing things that we may never see again_ My advice to you, kind stranger, is to treat everyone, the beautiful and the broken, as flowers that come and pass away. You can’t tell just by looking how long they’ll last. So cherish them while they bloom, even if you’re just passing by. * * * 1. Even though I do it, too. ↩︎ 2. I always heard it as _pay-paradise and pudapuparkinglot._ I had no idea what that line meant, but the lines before were enough. ↩︎

To me, not having access to a good dictionary feels a bit like forgetting to put on clothes. When Flaubert throws a big word at me, I marinate in my own confusion, helpless and vulnerable. Those of you who rarely disconnect from the internet might be surprised to learn that the dictionaries for many desktop dictionary apps live in the internet, not in the app, and stop working when offline. This seems so silly! The biggest dictionary in the world would be smaller than my `node_modules`. Since paper dictionaries are passé, and rarely have the word I’m looking for, when travelling, I tend to turn to my Kobo dictionary even if I’m at my computer. It rarely has the word I’m looking for, but at least it’s quick to find out that it doesn’t have it. After going through the rigamarole of selecting and adding dictionaries to KOReader, I thought, how hard can it be to use one of these on Linux? Let’s just say: this is my third attempt. As with most things Linux, it’s only hard if you don’t know how to do it. And even the first time, it isn’t hard, per se, it’s just annoying, and confusing, and barely worth it. ### INTERLUDE Have I mentioned that I don’t like copyright? Anyway: I don’t like copyright. Ever since I read Cory Doctorow’s _Content,_ and probably long before that, I have disliked copyright. My main activity in life is creating this this tasteless, odorless material called “Intellectual Property”, and then, generally, giving it away. The reason configuring a good offline dictionary on Linux is hard is, you guessed it: _copyright_. If there were _good_ dictionaries that were public domain, distributions would bundle them. As it is, there’s Wiktionary and Wordnet, and the excellent but hopelessly out of date Websters 1913. If you want a really good dictionary — say, a comprehensive Oxford — you have to bring it yourself. You have to decide which one you want, and then buy it in the right format, or download inscrutable, volunteer-converted, possibly virus-riddled `dict.dz` and `.idx` files of uncertain copyright status from an obscure MobileRead post. The contenders: * gnome-dictionary (online by default, but supports dictd servers) * dict (command line, online by default) and dictd * stardict (GUI) and sdcv (command line) * GoldenDict * Artha * Wordbook (untested) If you are content with Wordnet, `sudo apt install artha` and you’re done. (N.B. Artha is technically a thesaurus.) It has some cool features like hotkey lookup that just works, and regular expression search. Or, the less attractive but more powerful GoldenDict: sudo apt install goldendict goldendict-wordnet I am not content with Wordnet. I am also not content with a GUI only app, because part of the motivation for this project was that I wanted to export my Kobo vocabulary builder table, and generate an Anki deck of flashcards with the words I had saved on one side, and a good definition on the other. You can easily add bought/pirated/out-of-copyright dictionaries to GoldenDict (Edit > Dictionaries, and click add), and it seems to take the same format as KOReader. ### Command Line Usage To get sdcv working, you need to install it (`sudo apt install sdcv`), and add dictionaries to `/usr/share/stardict/dic`. Or, go where your dictionaries are, and run: ln -s /path/to/dictionaries/this-is-the-dictionary-i-want/* /usr/share/stardict/dic Note that linking to top-level folder into `dic` doesn’t work: it needs the individual files in there. You can then run `sdcv --list-dicts` to check that they added successfully. You can also use a specific dictionary in future commands, using what shows up here. My next surprise was that `sdcv` output raw HTML on the command line. This is probably because I’m using the wrong conversion, but my bandaid solution was to pipe it through pandoc, and into a pager (since good dictionaries tend to have exhaustively long entries): sdcv --use-dict "this-is-the-dictionary-i-want" test | pandoc -f html -t plain | less ### Hotkey Lookup with GoldenDict There was one final feature that I wanted: an equivalent of long-press lookup on ereaders. GoldenDict probably has this built-in, but I didn’t get it working. Instead, I added a custom keyboard shortcut, with the following contents: bash -c "goldendict \"$(xsel | tr '\n' ' ' | sed -r 's/^[^[:alpha:]]*([-[:alpha:]]*).*$/\1/')\"" Don’t ask me _how_ it works: it works, and that is enough. ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series.

I often need to get the internal UUID for a Ghost CMS post: for example, in order to save it as markdown. Previously, I would go to the post, append `/edit` to the URL, hit enter, and then copy the last section of the URL, which is the UUID for the post in the editor. Tedious! As I suspected, the post ID is right there in the HTML of the post: it’s used to load comments for the right post. So, like I did for appending `/edit` to the URL: Drag the link below to your bookmarks bar and click it when you’re viewing a post on your Ghost blog: Edit Post! It will append `/edit` to the url and reload, so you can edit your post — Built with ❤ by @curiositry (Twitter, GitHub) .... and linking to text fragments, and jumping between staging and production versions of a site, I wrote a two-line bookmarklet to extract the current post ID, and put it on the clipboard: let postId = document.querySelector("[data-post-id]").getAttribute("data-post-id"); navigator.clipboard.writeText(postId).then(() => { console.log('Post ID copied to clipboard:', postId); }); Drag the link below to your bookmarks bar, and click on it when you want to get a post ID from your blog: Copy Ghost Post ID to Clipboard (Thanks to Chris Zarate’s for making the excellent bookmarklet encoder applet that I use every time I need to make a bookmarklet. It works great, and can even be saved and used offline.)

In my travel & packing notes for next time, there is a line item — slightly above the assertions that “the eggs are usually gross” at hotel breakfasts, and “ferry WiFi is varely[sic] worth using” — that says: _I rarely have time/energy to do as many things as I think_. I travel a reasonable amount, and tend to feel overbooked leading up to a trip, and think, _oh, I’ll have time for that on the drive/plane/train/ferry._ So I stuff my backpack full of torn-out notebook sheets of things to finish, and then sit in a bleary-eyed torpor eating food of questionable quality and staring out the window, while the things I brought to work on get crumpled and food-strained, and prepare to silently judge me when I unpack upon arrival. I have often pondered: why does time traveling go by so quickly? Why are those long hours so often effectively a write-off? The obvious factors are obvious. When traveling, I tend to be tired, potentially nauseous, hot, sweaty, greasy, frequently-interrupted, dehydrated, full-bladdered, cramped, incessantly-talked at, interested in the scenery, and saving my energy for whatever I am traveling to do. Eventually, I realized these facts, and made two lists: 1. _Fun or useful things to do in the van on long drives.md_ 2. _Things that it’s possible to productively work on when very tired.md_ But there’s something _else_ going on. Yes, I tend to be tired, sweaty, and full of junk food. Yet, even when I am a passenger, I don’t even tend to get around to doing _simple_ , non-intellectually-taxing tasks! I noticed that there is one thing I can reliably do: _read._ Specifically, read on an e-reader. Often, I can read reasonably taxing books; almost always, I can manage to read something light, at least for a few hours until my eyes tire. It was by thinking about this one thing that I reliably get around to doing when travelling that I discovered the hidden gumption trap that was catching the feeble productive impulses my brain and body do occasionally produce. I call it _**backpack activation energy.**_ What is it that makes e-reading easy? Three things: 1. It only requires one item, which has endless battery life and a built in light, and which is easy to hold even when cramped. 2. My Kobo is 7.6 mm thick, lives in the most accessible place in my backpack, and can reliably be extracted and returned, blindfolded, no matter how overstuffed my luggage is. 3. I like doing it. Using my computer is probably the second-easiest thing to get around to when travelling: it shares many of the same features. Reading physical books that I brought with rarely ever occurs, even though I doggedly persist in bringing them, and in theory reading paper books has a lot in common with reading on an e-reader. The difference? Paper books get banged up when traveling, and produce a peculiar fingernails-on-chalkboard sensation when I reach into my snake-digesting-an-elephant backpack and pull out my book, and it feels like tug-of-war between me and the six USB cables caught between the pages. Yes, you have just read an entire essay about the fact that **I don’t like getting things out of my backpack.** Finding and extracting things from a full backpack takes physical and psychic energy that I don’t have — unless, of course, it’s chocolate. What about solutions? I don’t have any good ones, but here are some that are slightly better than nothing: 1. **Use ziploc bags**. I put all small items that I use frequently in a few large clear ziploc bags in the main compartment, categorized by category (though categorizing by colour would look cool). Looking like a homeless person = in this case, totally worth it. 2. **Don’t bring so much stuff.** Last I checked, personality surgery hadn’t been invented yet, but I’m definitely on the waiting list. 3. **Partially accept partial defeat** , and then doggedly demand partial victory. Realize that BAE is a _thing_ that must be worked with. Then, figure out which productive things are just at or below your personal BAE threshold, and do those things, rather than the things for which there is no hope. 4. **Keep a notebook and pen in your pocket at all times,** so you don’t have to bargain with yourself{rel="noreferrer"}. 5. **Pretend you are wrestling a bull to the ground** when digging for your earplugs, and the sweating and grunting is part of the glory. 6. **Drink more coffee.** Sorry, it’s the best I’ve got. ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series.

I have been using Linux since before I owned my own computer, and for much of that time, have been comfortable with the command line (I'm a computer programmer, after all; though not a particularly fancy one). I have always been annoyed that Ctrl+Z doesn’t work in terminal (not to mention, Ctrl+S, and all other table-stakes keybindings. Keybindings should be universal, man!). I knew, vaguely, what it did. It didn't sound interesting to me. I slowly warmed up to GNU readline keybindings. They’re not as good as universal keybindings, but I sure miss them when they're gone, and now I've started using them when I write TUI dashboards. And I spend my time in a console text editor, in a non-standard shell, in tmux. Which is just to say, I'm not a _complete_ rube. But, job control? “ _Wuts job control?_ ” said the self-styled CLI enthusiast. I knew, vaguely, that jobs could be backgrounded (whatever that meant), but I frankly didn't see the point. If you need to run something in the background, just use `command &` like a normal person (and then watch it rudely spit characters into your terminal when it feels like it anyhow). So. Then I was working on a project in which I spent large swaths of my day SSH'ed into an underpowered box running way to many heavy processes, via a bastion server. The nature of the work required binocular vision, so — prepare for the yuck — I would split my home Tmux into two panes, open _two_ SSH sessions, knowing vaguely how gross that was, and double-hop to the server, using one pane for my main work, and the other for poking around the file system as needed. Probably a month into the project, which involved running interminable Python analysis scripts against a 20gb SQLite DB, I (_re?)_ read the wonderful post Shell Tricks That Actually Make Life Easier (And Save Your Sanity). There was that little `fg` thing again... Suddenly, I was able to run _one_ SSH session. Mind blown! It works like this: 1. Start long running process: `python3 my_epic_analysis_script.py --do-all-the-extra-stuff-too` 2. Background it with Ctrl+z 3. Do other stuff, like poking around the file system, or madly trying to recover from swap thrashing 4. When ready to check if it finished (we're not fancy enough to have it notify), run `fg` to bring it back This is amazing right on its own, but it gets even better. Did you know that modern CPUs with modern CLIs can run _two programs at once?!_ 1. Start long running process: `python3 oh_my_gosh_this_is_going_to_take_all_night.py --auto-resume-when-you-crash` 2. Background it with Ctrl+z 3. Start _another_ thing that’s super slow and resource hungry: `emacs` 4. Background that too with Ctrl+z 5. Where did all my stuff go? Someone tidied up my room and now I can't find the clothes I left on the floor! That’s a job for `jobs` 6. Which one do we want? The first one, or the second one? The first one: `fg %1` (← unclear what the percentage sign up to but there it is) 7. Still not done! Ctrl+z, then `fg` (we can skip the `%2` because there is no third job at present), then Ctrl+z, then, in the morning, `fg %1`. I sometimes wonder if someone who really _knew_ Linux would ever need to install applications. The problem, often, isn’t that the feature isn’t there: the problem is _discoverability._ It was the same thing when I was learning Vim. I would install all kinds of glorious plugins, and then, months or years later, learn that the base install provided 98% of the functionality I needed, it was it just was a bit weird, and hard to find. Now you can press ctrl+z on this blog, and `fg` on whatever you were trying to do when you found it. ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series.

****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series. I have a reputation for ~~laziness~~ efficiency, and have always considered colanders to be a waste of resources. Unless you need your pasta totally dry (in which case you have other problems), draining directly from the pot works fine. So that’s what I've always done. However, if you're not careful, it's easy to get a steam burn; and I think it's these preventable steam burns that drives the world's obsession with colanders. The first time I learned this trick for preventing steam burns, I was immensely pleased, and I've used it every time I drain pasta since. To drain pasta from the pot, you have to hold the lid on. Sophisticates use dual hotpads (or dish towels) to clamp each side of the askew lid. This is a safe and reliable method that has the same issues as colanders: waste and complexity. The fully grug-brained pasta draining method uses just hands, and pot. The problem: the lid has to be held on, askew, while the water is drained; the handle of the lid is generally in the middle; as the water is draining, it is pouring out, near boiling, inches below the hand holding the lid; steam rises from the water pouring out of the pot and burns the hand holding the lid. (This applies both to long-handled pots, and to pasta pots with two handles, in which case one gets hooked over the sink to hold the pot in place, and the other is used to hold the pot. With pasta pots, the steam can get _either_ hand, because it will also come out the gap where the lid is askew at the top) The solution is beautifully simple: blow the steam away from your hand. As long as you pour the water out reasonably slowly, a steady stream of air passing over your hand (don’t faint!) will keep the temperature pleasant, no hot-mitts or colanders needed.

2026-05-18 On May 17th, a trader bet ~$20 at 1c per share (implied 1% probability) on the wrong bucket for Mexico City temperature, and then bet again. That bucket won, and ~$50 turned into ~$5000. On May 17th, Polymarket’s high temperature market for Miami was missing the winning bucket, and the ones around it. Then, the May 17th market for Miami was “archived” — which is not what happens to markets that are resolved normally — along with the money that people had bet on it. Other markets followed (or preceded; its hard to find a clear timeline). There was no official response from Polymarket about what had happened, and speculation was confined to the comments, X, and Reddit. Someone speculated that it was an exploit, pointing to the Mexico City trader’s wallet (though that person might have just noticed something was weird, or gotten lucky): > 🚨 Bad resolution on Polymarket > > Miami May 17 Highest Temp (86-87°F) resolved as No ❌ > > But official high is 87°F on Wunderground KMIA. Resolved way too early.@Polymarket @UMAProtocol can we get a review?#Polymarket #ResolutionIssuehttps://t.co/Ml6wUxzieC > > — 𝔻𝕖𝕗𝕝𝕒𝕩 💥 (@Deflax_) May 17, 2026 Rumors surfaced of an Oracle bug: > POLYMARKET DIGEST — 18 MAY 2026 > > ⚠️ Bugs & Issues > Major weather market oracle failure caused wrong temperature brackets to be resolved across multiple cities (Miami, Mexico City, Seoul, Hong Kong). Mass archiving of weather markets led to positions disappearing from portfolios.… > > — Digest Polymarket & UMA Discord (@predictdigest) May 18, 2026 Eventually, _all_ US and international temperature markets disappeared, not just for May 17th, retroactively, but also for May 18th to May 22nd. High Temp Odds & Predictions 2026 | PolymarketExplore live High Temp prediction markets as of May 18, 2026. Track real-time odds and trade on The World’s Largest Prediction Market™Polymarket The only acknowledgement from Polymarket so far seems to be this tweet from @kea1on: > There has been a misconfiguration on a very small subset of markets - mostly weather markets, and a couple of tweet markets. We have archived all the misconfigured ones and are going to issue refunds for these markets. > > — keaton (@kea1on) May 18, 2026 The Polymarket status page shows no notices for May 16th, 17th, or 18th: Polymarket - StatusPolymarket StatusPolymarket I don’t know what's going on, but I thought it was strange enough that someone should collect what facts are available, and none of the big news outlets seem to be covering it yet.

On the stock market, prediction markets, and elsewhere, things don't have One True Price (or, therefor, one true valuation). There is the price people with the thing are willing to sell it for, and there is the price people without the thing are willing to pay to get it. These two prices are generally not the same, and the gap between them — the bid–ask spread — is often wide. After decades of considering myself a fairly canny bargain hunter and negotiator, I noticed something: if I bought something that was “an incredible deal”, and then tried to sell it, I generally lost money. What?! Isn't the idea that something is a "good deal" if the asking price is below its value? And if its asking price is below its value, doesn't that mean that you should be able to sell it at its value? _No no no!_ All this time, my mental model of the pricing dynamics at play was completely wrong. Because **the bid–ask spread is everywhere!** The bid–ask spread _generally_ represents the cost of doing business. In the case of the stock market, transaction fees (and liquidity). In-person used-good transactions, though generally fee-less, have an invisible cost of doing business. This, along with the fact that both buyers and sellers are opportunistic (and negotiation room is often included in the price), explains the bid–ask spread. The cost of doing business — the hidden transaction fee — includes the risk of getting scammed, or discovering that the item is worth significantly less than expected, and the time and energy of arranging to meet a scary stranger somewhere safe, getting cash out of the bank, asking friends whether you should buy it, researching whether it's a good deal, waiting in a coffee shop with sweaty palms waiting for the scary stranger to show up, etc. All this blithering is leading up to a very simple idea. Well, three ideas. **First:** good deals are expensive, not in money, but in time/risk/stress. There is almost always a reason something is underpriced. Sometimes, the reason is simply, sometimes things are underpriced for no reason. But the reason you _found_ the underpriced thing is because you'd been obsessively checking for underpriced items in category X every morning, instead using that time to making enough money to buy a fairly priced item in category X, or decide you don't need the thing in the first place, and can re-assign the time to reading Proust. **Second** : possessions are far more expensive than they seem, when you include the deferred financial and attentional expense of maintenance and disposal. I highly recommend David Cain’s exploration of this idea: Everything Must Be Paid for TwiceOne financial lesson they should teach in school is that most of the things we buy have to be paid for twice. There’…Raptitude.comDavid Cain **Third:** If you want a good deal, don't think about whether you’re getting an incredible deal._****_(You will usually answer, _Yes!)_ Instead, ask: how much could I reasonably get for this if I immediately flipped it? **How much would _I_ be comfortable charging?** (For bonus points, _how much would I charge a friend if I sold them this_ , or _what would I let someone negotiate me down to if I was selling this._) This flips the set-point from **is this a fair ask-price __** to **is this a fair _bid_ -price. **Which can turn something from a great deal into a pretty bad deal! If you're just getting something for yourself, and you're in a hurry, it might not matter. But if you're getting something, and part of your plan is that you want to be able to sell it, at a minimal loss, if it turns out buying it was a mistake, then you shouldn't pretend you'd be able to sell it for the price you'd be willing to pay for it.

Suppose one is thrown from one’s horse. There is a period, perhaps, of being dragged behind the horse, with one foot stuck in the stirrup. Then, a period of sitting on the ground in a cloud of dust, feeling dazed and disoriented. Then, perhaps, a period of feeling one’s head, and moving each finger and toe to make sure they are still attached, in preparation for standing up, with a grunt. Then one must look for ones horse, shielding the eyes from the sun, and squinting into the distance. There. The pursuit of the horse begins. Once the saddle is regained, there may be a period where the rider is clinging on, canted slightly to one side, but hardly in control: wide-eyed and galloping, it may occur to the rider that the scene is ridiculous, and, specifically, that the scene makes _them_ look ridiculous, and, in the interest of propriety, it would be best to let go of the whole idea of getting back in the saddle, and instead, to accept defeat, and walk, but walk in dignity. My point is: dignity, at times, is less important than clinging on, and continuing to go in the direction you were heading, even it creates, briefly, a source of amusement for passers-by. Re-starting a habit feels a bit like trying to get back on a horse that’s run away from under you. It’s an awkward process. To succeed, you must try, and trying doesn’t always look hip. Fortunately, looking hip is optional. ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series.

It's hard to sneak up on yourself, but worth it. It's not really as paradoxical as it sounds. It isn't your _whole_ selfout sneaking up on your whole self. It's more like a conspiracy in which some parts of yourself collude against other parts of yourself. Or, better yet, a theater piece: the part being fooled _knows_ its being fooled, but lets it happen anyway. I was reminded of the joys of sneaking up on my goals when I was out for a long run in the drenching rain and snow yesterday. The half-way point was a coffee shop. We'd taken a back route that I'd only been on once before, which I remembered taking us _directly_ into the lap of luxury. I was timing my catastrophic indigestion to hit the moment we arrived at safety. But, lo! The back road popped out only about _halfway_ to the coffee shop. The whole time I had been basing my survival on a delusion, and it worked great! The coffee shop was _just_ around the corner for twice as long as it should have been, and as a result, I was able to keep up the threshold-run pace, and made it to the washroom in the nick of time. Pretty much every single time I have broken my all-time running distance record, it has been by a sneak attack. I am "officially" only going to go long-ish. But a part of my mind knows its a good opportunity to break my all-time record. It makes sure to tuck a few extra energy bars into my pockets or hydration pack. There's nothing to be scared of: there's no commitment to doing something hard. There's just the opportunity, and the preparation to make use of it. But guess what: at kilometer 25, I don't _have_ to run a marathon. I just have to run another five, to get to a nice round 30. And then at 30, might as well make it 35. And at that point, 42 is just around the corner, so it'd be a waste not to run the full distance. At which point _fifty_ is just around the corner. And on and on until legs give out. At no point do I contemplate running a marathon, until I already have. I only sucker myself into going just a _bit_ farther, and keep on doing that over and over again. And rather than realizing, _I keep on suckering myself into going farther than I planned! This is mission creep!_ I re-frame it, and think, _that wasn't so bad, why not throw on another 5K…_ The few times when I planned on breaking my distance record, some joint would develop the wrong kind of pain, or other responsibilities would get in the way, and I'd be bent out of shape about "failing" to achieve my goal. So I have become superstitious about using my preferred method, and sneaking up on my goals. First I cover my eyes, then I say boo. It works every time.

If I throw an unfamiliar word or concept at you in the first paragraph of this essay, will you bother to look it up? If the answer is yes, you can close the tab, unless you want to bookmark it in your “people saying things I already believe” folder, to inflict upon someone less advanced. I'm pretty good at bothering to look up _words_ I don't know. But what about when a blogger or AI model starts justifying their reasoning with little mathematical symbols, and it makes me uncomfortable, because I don't know what they mean, and am too proud to ask? Not so much. Now that I'm old and busy, I find it harder to get around to learning-for-the-sake of learning. It’s much easier to get around to learning in the process of doing! Yet, opportunities frequently pass me by. I scroll past the unfamiliar symbol; the elk footprints go uninvestigated; “that's funny...” is left at that's funny, rather than, “aha!”. Dan Luu calls this "willingness to look stupid". Slime Mold Time Mold call it the scientific virtue of stupidity. I call it bothering to understand. Slowing down, amid all the breakneck chaos, to figure out _why_ it crashed, rather than just restarting it. This is unlikely to pay off in the short term. But _next_ time it crashes, or maybe not even then, but next time something vaguely like it does something vaguely like what it did, I'll have a foundation to build upon. Knowledge compounds. You only have to make the wand once, then you can blow bubbles with it forever. Make the wand. ### Techniques for putting this principle into practice Principles are mainly useful when applied. Talking about them isn't the point. In this case, two things tend to get in the way, depending on context: discomfort, and inconvenience. #### Social discomfort * **The sooner I start bothering to understand, the less awkward.** This applies over the course of a lifetime, in relationships with specific people, and within an interaction. * It's less awkward to ask ignorant questions when you're young, than to always be an ignoramus * It's less awkward to learn how to pronounce / remember someone's name the first time you hear it * It's less awkward to clarify unfamiliar terms and concepts the moment they come up: if you have to go back and clarify them later, after _pretending_ to understand them, it calls your bluff, which doesn't look good. * **Just get over it** , it doesn't get easier, but in the long run, it's the right thing to do, _and_ higher status, to bother to understand, even if that means crossing a moat filled with alligators of (possibly imagined) social judgement #### Inconvenience Without the pressure of an in-person interaction where I would lose credibility or waste people's time by asking dumb questions, the main barriers become time, energy, and inconvenience. * The built-in, **long-press-to-look-up word in dictionary** on e-readers makes it convenient to learn unfamiliar vocabulary without breaking the flow of reading. The catch: unless you load a custom dictionary, if you have a tolerably large vocabulary, a significant percentage of unfamiliar words won't be in the default dictionary, especially if you read things heavy in neologisms and technical terms. * KOReader takes this a step further with built-in Wikipedia lookup * I **write down things to look up,** if I'm in the middle of something that I don't want to interrupt. And rather than typing my paper notes into a text file, I often type them directly into the Firefox search bar, even if I'm working offline. Then, next time I'm using my browser, the tabs will be in my way, and I won't forget. * **Doing things myself,** whether self-hosting a website, fixing a computer, or building a rock wall, results in learning all kinds of things by necessity and coincidence, and it's much more natural than taking a course. * If I'm **hanging out with an expert** (and almost everyone is an expert at something) — say, a radiographer scanning me at the hospital — I try to ask as many inquisitive questions about the fundamentals as I can get away with, early on. (It's easy to think, _that's a dumb question, I could just Google it._ But chances are I _won't_ Google it.) This way, I'm more likely to be able to have an interesting and educational conversation with them, and then an even better conversation with the _next_ radiologist I talk to. * I try to **get up and investigate first hand** when I encounter things I don't understand, like a weird beetle or a mysterious object. I have a natural resistance to getting involved with the physical world, so I make a point of getting out of my comfortable seat and going over to the thing to learn more — to get down on all fours, prod, sniff, or peer, as the circumstance dictates — even if it feels undignified and child-like, and gets my knees grubby. ###

### I: Curators Skim Curators and gatekeepers, because of the nature of their work, default to skimming. It's easy to be annoyed that the editor might only read the first page (or the first paragraph, or even the first sentence) of a story you spent days, months, or years writing. But most of the time, the editor only needs a few seconds to gauge the quality of the writing, and whether it’s a fit for their audience. _ This interesting dynamic sunk in after reading_ _The View from RSS_ _, a behind-the-scenes view of curation by the glamorous and shockingly intelligent Caroline Crampton, who describes what it’s like to live underneath a pile of two thousand RSS feeds, possibly including this one. Full disclosure: I am plotting to use the ideas in this article to get featured by The Browser again, preferably without having to spend 6 months writing_[ _a 5,500-word essay_ _to do so. ]_ Songwriters have shortened the average length of the instrumental intro, to get to the catchy chorus hook while the goldfish are still listening. The usual story is that attention spans are declining. But is it the _audience’s_ attention spans that are dictating this change? Or, is part of it a systemic bias for things that front-load the value, and are easily curatable? In some cases, there’s an incentive for curators to share things they haven’t deeply read: speed. If their heuristics are accurate, skimming-only allows them to be the first curator to break good articles, increasing their status as taste-makers. ### II: Skim Milking I realized underdog independent bloggers could use curators’ skimming against them. In the spirit of Gwern’s hacker mindset essay, we could optimize our writing _for_ these curators, by tailoring the parts that stand out to specifically catch curators' attention. Then I realized: this is what the popular writers do. Have you noticed all the James Clear-like life advice writers with their bold text, catchy statements, and soundbites? Is it a coincidence that the most popular writers use bold text and write eminently-skimmable articles? The boring takeaway: we could make our articles stand out to curators by being strategic about optimizing the things that the curators will hit as they skim. (The tricky part is how to deliver the best value for the end-reader at the same time as fishing for curators.) ### III: Applied Antimemetics If, for some reason, we wanted an article _not_ to go viral, ala Nadia Asparouhova's anti-memetic self-keeping secrets, we could deliberately emphasize the boring parts. The exact same content, with different emphasis, could make it so that curators who didn't read the whole thing would think it was a badly written, boring article. But readers who read the whole thing would realize that it wasn't, or even, that the parts that weren't in bold directly contradicted the parts that were in bold. In the spirit of adventure, I could publish two versions of this essay. One of them would highlight the interesting parts, and attempt to optimize for the skimming curators and gatekeepers. The second version would only highlight the worst and boringest parts, and optimize for the readers who read the whole thing, and specifically try to _prevent_ the curators from discovering it. ### IV: Holes in the Sieve Certain articles or books take a while to get rolling. Steinbeck comes to mind: an entire chapter about the weather. It seems these would be at risk of being passed over by a skimming curator. And some types of material resist shallow engagement. They seem boring, or dumb, or empty when you blaze though them. But, sometimes, they’re not! Once a week I go through my comparatively modest collection of ~151 RSS feeds (some of which are, in turn, curated collections of links), as well as a few aggregators, and select my week’s reading material. Like Caroline, “ _I’ve become very good over the years at allowing my eye to slide over everything, stopping when I see a headline or phrase that looks promising._ ” The first step of triage happens in my RSS reader. For noisy feeds, I only read the headline. Or, I skim the article. I open the articles that seem interesting, and I save the best to Pocket/Wallabag/Instapaper/Readeck. Arguments for and against things, self-help, and hot takes are easy to skim, and get a decent sense of whether they will be interesting. These are what I will call “point based” writing. They have a point, and it’s either right or wrong, and it’s either relevant or irrelevant to my life, and it’s either well written or not. But there are a few categories of things that can get triaged when they shouldn’t. **Buried nuggets.** Often, 98% of the value I get from an essay is a single idea or piece of information, which I haven’t encountered before, buried somewhere obscure — often in a footnote. Real-life example from yesterday: a Dwarkesh Patel interview with Gwern Branwen crossed my screen. I skimmed it briefly, and triaged it as “modafinil mode” content (the hyper-intellectual cottonwool prose of the Rationalists, which takes a lot of effort to understand for an uncertain payoff). Because the software I wrote for my ereader was buggy, the articles I chose for the week did _not_ show up on my ereader, and the offending 16,000-word Gwern essay downloaded itself, along with a bunch of Amazon book listings and other random crot. I was bored, so I read it, along with some of the other triaged essays. Some of it was modafinil mode, and I skimmed, but two little asides made it worth it. I learned that like me, Gwern is broke, even though he’s a household name. And I learned that he isn’t into psychedelics, which totally broke my model of Gwern as the reckless pharmacological self-experimenter. (His argument nearly matched an unpublished essay I wrote, _Experimentation and Downside Risk._) **Emergent payloads.** Another type of material that breaks triage workflows is when the whole is greater than the sum of the parts. The piece, every part of which is mundane, creates a _vibe_ , and something about the gestalt of all the boring subpieces is educational or moving. The Russian classics come to mind. **Quiet depth** is also triage-resistant. Some people, like Wendell Berry, write in a very plain tone about ordinary things. It might not jump out at you. They don’t use the flashy literary tricks, and the pace skews slow. But if you pay close attention, and let the seeds they plant in your mind bloom, something extraordinary happens. Often, these seeds appear to be platitudes. Alone, they would bounce off the modern consciousness. But the rest of the essay fertilizes the readers mind for the small, commonplace seed. If you rush through these essays, the seed will not germinate. It will just fall on dry soil, and get laughed at, because it’s such a tiny and ordinary seed. ### V: Dumpster Diving To deal with un-skimmable material, curators follow _other_ curators that keep tabs on specific beat. If a sub-curator with a good track record surfaces something that seems boring on the surface, chances are it’s worth a deeper look. However, even if the multi-leveled curatorial juggernaut worked _perfectly_ , there's another concern. The curators are trying to find the _best_ articles for their audience. The more influential the curator, generally, the more general the audience. Everything they pick has to be good, and it has to be evenly good: palatable enough they don't lose their audience. But you and I are not an audience. We are individuals, and our needs cannot be optimized for by a system of this scale. Curators are a band-pass filter. They remove most of the bad stuff, and some of the good stuff that can be confused for the bad stuff. But I have found some real treasures in dumpsters and at recycling centres. You can't always judge the best by the mean or the median. I have built electric bicycles out of windshield wiper motors, even though the entire car they were removed from was dead and smashed up. Curators don't tend to serve us smashed up cars with cool components that still work. That's why, in addition to _The Browser_ and _Hacker Newsletter,_ I subscribe to high-noise-to-signal uncurated feeds like _HN Blogs._ Sometimes I want to go to Tiffany’s; sometimes I want to hold my nose and go dumpster diving. The average is worse, but you find things you won't find anywhere else.

I often stay up late reading books. I am told this is called “Revenge Bedtime Procrastination”. Then I get up late and groggy, or early and under-slept. When I was reading an excellent old English novel last night, it occurred to me: these little habits of “wasting time” are simultaneously totally harmless (in a vacuum), and potentially _sometimes_ very consequential, in context. I spend a lot of time thinking about what _could_ have happened if what did happen didn't happen. And I have always felt a bit ripped off that we make trade-offs every moment of our lives, _without knowing what we are trading off_. Predicting the future seems, by and large, futile. But if there's any part of the future that we _might_ be able to predict, it's our own tomorrow morning. So I have taken to trying to calculate, any time I am contemplating harmlessly wasting time or staying up late, not, _would I be doing something important if I didn't do this_ (the answer is, usually, no), but, would I be doing something important _when I end up doing the thing that I would otherwise do now._ Because it's the second and third order effects, as usual, where the consequence shows up. If I stay up late reading, that might seem fine, if I have nothing to do the next morning. But if I have something to do early the morning _after_ that, I will be tired, because I will be on a late schedule. Or: if I read Hacker News for twenty minutes, that might seem fine, because I'm too tired to get any work done anyway. But slacking off isn't displacing work (which wouldn't have happened anyway, according to the narrator). It's displacing things like actually resting, and doing chores. And _those_ things, if displaced, might in turn displace work at a time when work _could_ have been done! In which case, in terms of causality, even though it seemed like it wasn't, reading Hacker News was a _direct_ tradeoff against accomplishing the most important item on my todo list. It is perhaps worth mentioning that I read until two AM last night. And I _may_ have watched two episodes of Star Trek this afternoon, because I had a cold. So here I am, at 11:30 PM, having accomplished approximately nothing in the day other than planting fifteen scarlet runner bean starts, with silence all around me except for the dishwasher, reflecting on the wisdom of my choices.

****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series. I'm looking for a new laptop, suddenly, which involves a lot of time with elaborate search criteria on Ebay/Craigslist, and visits to PassMark's CPU benchmark so frequent that I made a URL-bar shortcut for the site. From my experience of previous laptops, this is what I'm looking for, and why: **Main criteria:** 1. **Good Linux support.** I don't want to have to drag my developer laptop into the world of open source kicking and screaming. My experience so far: ThinkPads good, Macbooks bad, everything else mediocre. 1. This constraint narrows the field down to Lenovo ThinkPads, Dell XPS Developer Edition, Framework Laptops, and Linux-first startups and resellers such as System76, StarLabs, Slimbook, and Tuxedo. 2. **Durable and repairable.** Old ThinkPads are my benchmark. I have had to repair every laptop I have owned, myself, and I want it to be _nice._ I use my laptop hard, and travel a lot. 3. **High enough spec to run SOTA local models.** For me, this means ~64gb–128gb RAM, a late model gaming-grade GPU, and a CPU faster than what I had (Ryzen 7, ~20k CPU bench). <14B models don't cut it. I'd like to be able to run 70B. And I have spent a lot of time trying to jam overlarge models into limited RAM/VRAM, and determined that it's a waste of time. **Secondary criteria:** 1. **15+ inch screen.** I've mostly used 13–14 inch laptops, and I find they're too small to feel fully productive. 2. **Loaded to the gills with RAM and SSD storage** , since they're getting more expensive. I want my next laptop to have as much RAM and storage as I'm going to need from day one (2TB+ storage, and ~64gb+ RAM). 3. **Cheap**. Did I mention that I wanted my insanely powerful dream computer to be dirt cheap? There's no way to get these specs objectively cheap, but patience, bargain hunting, and sacrificing on cosmetics can get you part-way there, and close the gap between purchase price and resale value. 4. **Not a gaming laptop.** My last laptop was a gaming laptop, because that was the tradeoff for getting good specs at a good price. But I find gaming laptops ugly, and less care seems to go into the design and build quality, since they go obsolete so quickly. Since my gaming laptop broke after two years of use, I'm not getting another gaming laptop. 5. **RJ45 Ethernet Jack + 3.5mm headphone jack**. I don't want to have to use a USBC hub for things I use every day. **Nice to have:** * A 4k screen — but I'll settle for a good non 4k screen. (To go back to my old saw, the screen on T14s and Dell XPS are way better than the screen on my gaming laptop. The screen on my Asus G14 _sucked_.) * Ability to drive multiple external 4k monitors — but I'd settle for 1. * Easily-swappable harddrive + battery + RAM (old ThinkPads totally spoiled me) * Coreboot * Good battery life * (Open)BSD support * Non-janky keyboard and trackpad * Lots of USB C + USB A ports * A keyboard drain :) So what options are there that match these criteria? The ones that come to mind are: * The Framework 16 (expensive for the specs, but pretty much perfect otherwise) * ThinkPad T15, T16, or X1 carbon ← probably the sweet spot * Various of the (generally hard-to-find-used) StarLabs and System76 workstations I'm hope I'm missing some good options. If I am, let me know! I'm also curious, if you have vaguely similar criteria: what things _are_ worth sacrificing, in order to prioritize more important things? My strategy so far has been to buy dinged-up junkers, used, and then fix them. And to do my research, and then wait, and wait, and wait. If I billed for the time I spend trying to save money, I would be a rich man.

2026-03-09 Today, for no apparent reason, my 2021 Asus ROG Zephyrus G14 experienced a sudden power failure. This was the second or third time I had experienced abrupt restarts, and, apparently, the last. I have spent the last four or five hours tearing down the laptop trying to troubleshoot what went wrong, and in the process, have thought of several ways that I could have been better prepared for such an eventuality. In no particular order, here are some things that you may realize after a sudden loss of digital competence, and might want to realize _before_ your laptop shuts down for the last time: 1. Backing up is not a luxury, it's a necessity. _It's only a matter of time._ 2. There is probably a gap between how often you think you should back up, and how often you back up. Try to split the difference. 3. Hard-drives are expensive, but lack of backup hard drives is _more_ expensive. 4. A good back-up system doesn't just include (multiple, offsite, verified) back-ups. It also includes a spare laptop. And it also includes the piece I was missing: a way to get the drive from your primary laptop into your secondary laptop. 5. Multi-factor authentication is a key failure point, so make a plan. Personally, I think mandatory 2FA does more harm than good: The number of times that I have been slowed down at critical times (such as when fending off a spam attack), or fully locked out of my accounts for stupid reasons, far outweighs the number of times my data has been hacked because I didn't have 2FA enabled (~0). MFA is good, but if the narrative is that it's there for the _user's_ benefit, maybe check in with the user about their priorities? 6. Inspiring tangent: there are still people out there that are _incredibly hardcore_ about repairing laptops. Following ten-page troubleshooting threads on the badcaps forums, where a Fluke meter and thermal camera are table stakes, makes me feel like a Neanderthal with a soldering iron. This is a good thing. 7. Maybe don't buy a laptop that's famous for having the motherboard go bad, and has liquid metal thermal paste that _melts and runs out and short circuits the motherboard?_ 8. If you insist on buying this laptop, maybe learn about these weaknesses before you get all clever and dock your laptop on its side so you can run performance profile all the time with custom fan curves. 9. If you're wondering how to get around to daily-driving OpenBSD, catastrophic power failure on your Linux laptop is a solid option. ### Conclusion **One:** take a moment to be grateful for the fact that your computer works and you still have access to all your data. **Two:** Go back-up your computer, _right now._

When I was a kid, and my older siblings were working their way through the Royal Conservatory of Music curriculum, I thought notated music was called “sheep music”. [1] I was reminded of this fact last night, when I was reading Thackeray, and realized that for twenty odd years I have been hearing “wedding banns” as “wedding _bands”._ [2] The cause of both of these silly mistakes is the same: I’d never read the word, I’d only heard it spoken. The opposite problem, pronouncing book-learned words wrong, causes me even more frequent and severe embarrassment. How is a blissfully-politically-ignorant twenty-something to know that **Reuters** is pronounced “ROY-tərz” not “ROO-tərz”, because the British news agency was founded by a German baron? By saying “ _according to ROOTERS […]”_ , and getting corrected by someone who watches the news, that’s how. (I was relieved to learn that well-respected generalists, such as Gwern Branwen, have the same problem.) Even though I’ve been a member of the Tildeverse for 12 years, and type tildes countless times per day, I was recently corrected when I pronounced it “til-DEE”. Crestfallen, I looked it up, and Merriam-Webster agreed that the correct pronunciation was til-DEH. (Wikipedia lists both, which was comforting: my pronunciation wasn’t wrong, per se; just unpopular.) _ This is exacerbated by_ [_my tendency to chunk/pattern match words_) _rather than sounding them out. For many years, I inserted extra letters, so “Balsamic” vinegar became “Balsalamic” vinegar. It sounded similar enough that nobody corrected me. ]_ There is further confusion because people often try to “correct me” to the American pronunciation. (I pronounce data “DAH-ta”, which is common in Canada, Ireland, and Scotland, but most people pronounce it “DAY-ta”, so I have to chameleon sometimes.) **There isn’t enough overlap between what I read and what I hear people talk about.** I often have been reading about something for years, and writing about it, giving me time to think I know how to pronounce the word. Eventually, it comes up in casual conversation with someone who knows the word, and often it turns out I’m wrong. (Unfortunately, me excitedly telling my friends about things I read on the internet doesn’t help.) It’s the old _my Venn diagram looks like John Lennon’s glasses_ problem. Solutions time! The obvious answers: 1. **To address bookworm pronunciation syndrome** : use the dictionary audio pronunciation guide, or IPA (International Phonetic Alphabet). 2. **To address sheep music syndrome** : Look up new words I that learn orally. These are fairly bad answers. The problem isn’t that I _can’t figure out_ how to pronounce things. It’s that I _don’t know when I don’t know_ how to pronounce a word. It would be way too cumbersome to have to verify the pronunciation of every word I learn. A slightly better plan: 1. Make a point of using words I learned from books and essays in conversation, to give people a chance to correct me. And maybe even record podcasts of me garburating the English language, to increase the chances of being corrected. 2. Make a point of consuming at least _some_ technical and scientific content via audio and video (ick) rather than reading, so that I can cross-reference pronunciations. As it is, I don’t really listen to podcasts, and other than Khan Academy, rarely listen to lectures. Maybe binging on some 3Blue1Brown would help. 3. Work through a deck of IPA (International Phonetic Alphabet) Anki flashcards (this one is the best I’ve found), so that I can get the full benefit from dictionary definitions when I use them. If you have found a better solution to bookworm pronunciation syndrome, send me an email, or let me know in the comments. * * * 1. I thought this was strange and funny, but the world is pretty strange and funny, so this particular strangeness didn’t seem remarkable. ↩︎ 2. … which _sort_ of makes sense, since there are various traditions of binding the bride and groom’s hands together? Also, I was thrown off the scent by the fact that wedding rings are sometimes called wedding bands. ↩︎ ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series.

During the review phase of a website migration that involved decades of content and a complex URL structure, which meant going through the site page-by-page checking for content and formatting that was missed or mangled in the migration, I made two simple bookmarklets: **Go to this page on old {{client_name}} site** and **Go to this page on new {{client_name}} site**. They didn’t replace side-by-side comparison and automated diffs, but I found these bookmarklets surprisingly useful for quickly tabbing to the same page on the other version of the site any time I noticed something anomalous. The code was ridiculously simple: javascript:void function(){javascript:(function(){window.location=window.location.toString().replace(/www.oldsite.com/,"www.newsite.com")})()}(); Here is a hand-coded interactive widget, if you’d like to make your own _old site, new site_ bookmarklets for a migration: Go to page on old site Go to page on new site Toggle to page on other site Put the URLs of the old site, and the new/staging site, into the two inputs above. (Skip the protocol, but include the subdomain if there is one.) Then, drag the gaudy bookmarklet buttons to your bookmarks bar. ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series.

You can sort of have a conversation with a book. I sometimes write rude comments in the margins when I disagree with the author. But the Stoics recommend that when you’re insulted, you just ignore it. Books are incredibly advanced stoics in this regard. And it’s kind of infuriating to insult someone and have them not even notice. Conversations with books are, generally, parasocial. I’d like to be able to have interesting ideas on my own, but as I get older, the more I find that this doesn’t happen. Ideas rarely spring up from within me, like a seep in the desert. More often, they become visible in contrast to something else. I realize I believe something when I meet someone who believes the opposite thing. Until then my belief was just the water I swam in. I realize I stand for something when somebody tramples that thing before my eyes. Or, ideas are drawn out of me. Someone asks a question, and I say something that I didn’t know I knew, but which seems to be true. When I just read and think and write, that works up to a point. I am exposed to new ideas. They combine and interact with the things I already know. I write about the synthesis, or defend one idea or the other. But it’s easy to get stuck following the same old intellectual cart-tracks, wearing them deeper with each trip. It might sound like I’m leading up to the idea that self-styled intellectuals should get out of the house and experience the world, whatever that means. No, no, no! I mean, that’s good too, but leaving the house is a lot to ask of a self-styled intellectual. A more realistic step: write really long emails to people who are at least _slightly_ different from you. Long conversations, whether in person or by email, draw out things I haven’t consciously thought of, but, it seems, have been mulling over in some background process for a long time. I am often surprised to discover what I think, in the process of replying to a question. Good questions can exhaust the cache of shallow opinion, and get to things that are more deeply held and interesting, and then, after a surprisingly small number of whys, to all the things I don’t know. And that’s like a quarry for new ideas. ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series.

Blogs are back. RSS is cool again. _(Said the blogger to his RSS feed.)_ Here is a collection of small, independent directories of blogs, personal websites, and smol/indie/slow/human web sites that I’ve run across and bookmarked. Most of them accept submissions. Some of them have RSS feeds. ### Directory-ish Directories ooh.directory: a place to find good blogs that interest youA collection of 2,388 blogs about every topicMataroa CollectionBlogScroll - Personal Blog & Site DirectoryAn open directory of personal sites and blogs, maintained entirely on GitHub.Personal Blog & Site DirectoryYe Olde Blogroll - Because blogs are the soul of the webYou are viewing a humanly curated list fine personal & independent blogs and sites that are updated regularly. No algorithms ever!BlogrollBlogroll ClubA home for everyone’s blog and post rolls.Blogroll Clubblogs.hnA tiny tech blog directory.Taylor TroeshIndieseek.xyz Indie Web DirectoryA web directory of Indieweb | Indie Web or Independent websites. We index the creative web.Indieseek.xyz Indie Web DirectoryIndieseek.xyz Indie Web DirectoryThe Big List of Personal WebsitesThis is a big list of links to personal websites.MinifeedMinifeed⬤ minifeedJamboreeFind pages and sites from all around the World Wide Web! Jamboree provides a directory of links to help you navigate the Web.PersonalSit.esPersonal sites are sick as hell, so this site was built so we can all discover each other’s. This directory of links are by folks that want to share their site with the world.Yes we got hot and fresh sitesHN Personal Websites Directory ### RSS-centric powRSS - Discover the Indie WebpowRSS is a public RSS feed aggregator for small and independent websites.Discover the Indie WebfiercefreesiaRSS.Socialfeedle: Search and Discover Quality RSS Feeds from Thousands of Blogs and PodcastsA search engine for blogs and podcasts, where every search is an RSS feed.feedleScourScour interesting reads from noisy feeds you can’t keep up with and smaller sites you didn’t know to check.ScourExplore Indie Blogs | Blogs Are BackBrowse our hand-curated directory of independent blogs. Writers on tech, design, personal essays, creativity, and more.Blogs Are BackBlogs Are Back ### StumbleUpon-like Read Something InterestingCelebrating great writing on the webReadSomethingInteresting.comKagi Small WebDiscover the small web - personal blogs, independent YouTube channels, and webcomics from genuine humans on the internet.The ForestRediscover the joy of getting lost on the webThe Forest ### Constraint-based The 250kb ClubAn exclusive membership for web pages presenting themselves in no more than 250kb.koehr512KB ClubThe 512KB Club is an exclusive list of web pages weighing less than 512 kilobytes.512KB ClubKev Quirk1MB Club: A collection of web pages weighing in less than 1MBAn exclusive membership for web pages weighing less than 1 megabyte.Hall of glory | The no-JS.clubThe no-JS.club ### Search Engines Wiby - Search Engine for the Classic WebWiby is a search engine for older style pages, lightweight and based on a subject of interest. Building a web more reminiscent of the early internet.Search Engine for the Classic WebMarginalia SearchMarginalia Search is a small independent do-it-yourself search engine for surprising but content-rich websites that never ask you to accept cookies or subscribe to newsletters. The goal is to bring you the sort of grass fed, free range HTML your grandma used to write.Marginalia Search

The big arguments for and against AI have been endlessly discussed, and I don’t feel I have much to add. AGI and existential risk; human obsolescence; power use; cybersecurity; safety + censorship; slop; misinformation. Also, I’m kind of tired of everything being about AI, which is why I have specifically avoided writing about it. But here is a list of petty grievances with AI, that don’t make the cover of WIRED. * Basically none of it is actually open source. Lots of the tooling is, but “open weights” is fundamentally different, and worse, than source available, let alone open source. There can be targeted attacks, or blatant censorship, in open weights models. They’re a black box: they aren’t _safe_. And _none_ of the state-of-the-art models release their code + training data. * This is not just big companies are bad: it’s partly because the training data is huge (and usually pirated copyrighted material, and too poorly vetted to want to make public). * AI is centralized even though that’s bad architecture. Check what server you’re connecting to when working with local LLMs: it’s invariably HuggingFace. But models are huge, and though there would be certain legal and technical hurdles, BitTorrent would be a far more efficient technology to distribute open weight models. * AI makes Nvidia rich, and I don’t like Nvidia because their Linux support sucks ♥ * Because it’s so resource intensive, AI is even _more_ of a Matthew effect than technology in general. In the local LLM world, it’s the people who can afford a late-model Macbook with 64–128gb RAM and a hefty GPU that get to use the actually good models while maintaining their sovereignty, so the people who are empowered become more empowered. This gets even worse the farther up the chain: the capital requirements mean they are only a handful of frontier AI companies worldwide. * AI is fundamentally non-deterministic. At societal and epistemic levels, this is, obviously, disastrous. AI has no conception of truth: only probability of seeing it in the training data. Non-determinism can’t be “aligned” away with RLHF: it’s baked in. Tesseract is vastly worse than AI OCR. But the damage a Tesseract error can cause is bounded. The damage of what AI can hallucinate is practically unbounded. This means some of the practical advantages of AI are offset by how carefully the transcript/code/OCR must be reviewed if it’s being used in a context where the truth/meaning matters. * AI’s mistakes are less obvious. Surface-level mistakes are a red flag in human output. AI makes fewer surface level mistakes, but more fundamental errors. Because our heuristics are trained on human outputs, this makes it seem more trustworthy than it is. * AI adds another layer between humans and the world, distancing them from the consequences of their choices. People spend more money when they use a credit card than when they use cash. Cash is an abstraction layer on work. Credit cards are an abstraction layer on an abstraction layer, making it even _more_ convenient to spend money. I worry that this distancing will make (for example) waging war with semi-autonomous weapons, or just trading on the stock market, feel even more like a video game than it currently does, and buffer the operator from the real-world consequences of their operations. Anyone who has read _Ender’s Game_ knows that this kind of gamification can end badly. * AI feels grievously inefficient. It took 29.29 minutes to OCR the 4-page handwritten draft of this essay with Qwen3-VL:8B. I didn’t get exact power draw, but it was likely at least 45 watts, and could have been up to the rated 110 watts TPD. A human brain is estimated to draw ~20 watts, and could do the task in a fraction of the time. This feels...wasteful. And this is just inference! (To be fair, humans require training, also.) * Probably, much of this will be optimized away. And part of it is inherent to general-purpose systems, which are, by definition, not optimized for a given task. After all, many operations that were once too inefficient for widespread use — full disk encryption, VPNs — are now widespread. * Without knowing what consciousness is, we won’t know whether AI has become conscious. We also don’t know whether whether AI is conscious will matter. Which is a scarier: conscious AGI, or AGI _without_ consciousness? * AI makes me feel dumb. ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series. *[AGI]: Artificial General Intelligence *[RLHF]: Reinforcement learning from human feedback

It’s unclear how long human thought and human writing will remain relevant, so we are growing hay while the sun shines, and before the autonomous tractors make such efforts seem quaint. Here are 25 new human-written posts, fresh this month, only _some_ of which are about AI. _For those who are new: in 2026, we took on a writing challenge that’s a departure from usual Autodidacts deep dives. This newsletter is the Q2 report, covering posts written in February (posts 26–50-ish of 100)._ If you’re in a hurry: my favourites are probably _In Praise of Earnestness_ __(not necessarily the most well-written, but something I feel strongly about) and _Sonder is a word I like_ __(more experimental style, and another topic I’ve been thinking about a lot). I enjoyed writing about not being fake in social interactions (and helping other people escape from their tapeloops), semi-accidentally improving my memory, feeling more creative in the evening, the benefits of simplicity, bothering to eat flax and do pushups when I feel like a wet rag, and making my chronic insomnia _productive._ A number of posts are on using local AI for optical character recognition and dictation: boring if you’re not interested in that, but quite a breakthrough in my workflow. You’ll also find assorted tech tutorials, and a few additions to my read-it-later migration saga (including a suggestion to use your nose as a touchscreen stylus). Here are this month’s posts! My Silver price API now has an RSS feed with the day’s bid priceAlso, I now know about RFC-822 dates, which is not something I was interested in learning aboutThe AutodidactsCuriositryA basic way to hotreload an AppImageRepack and run every time you hit saveThe AutodidactsCuriositryScript: bulk star Wallabag entries by URLUseful if you imported from Pocket CSV, and it didn’t include your favouritesThe AutodidactsCuriositryScript: bulk star Readeck entries by URL (with URL cleaning)A better version of the script I wrote for WallabagThe AutodidactsCuriositryIn Praise of EarnestnessLet’s make it cool again, because it is.The AutodidactsCuriositryAn extremely silly bug🐞The AutodidactsCuriositryOCR Typewritten Documents with a Local Vision Model (Qwen3-VL:8B + Ollama)It’s now possible to get results better than Tesseract, without relying on cloud servicesThe AutodidactsCuriositryHow to pause and resume a resource-intensive process on Linux with SIGSTOP: specifically, Ollama LLM inferenceKeep your system resources to yourself when you need them, and put them to good use when you don’tThe AutodidactsCuriositrySomething good happened to my memory, and I have a theory about what caused itAccidental strength training for the hippocampusThe AutodidactsCuriositryStar rating support has been merged into the Readeck KOReader pluginI ported over the patch I wrote for the Wallabag plugin, and improved itThe AutodidactsCuriositryHow to increase Ollama context lengthOLLAMA_CONTEXT_LENGTH environment variable didn’t have an effect, but there’s another wayThe AutodidactsCuriositryInsomnia isn’t a waste of timeSub-clinical insomnia is endemic among the people I know. And I’ve rarely heard anyone say anything good about it. We all know how important good sleep is for good health, and about the increase in heart attacks when daylight savings time kicks in, and that getting less than 6hrsThe AutodidactsCuriositryMaybe don’t optimize conversion rate?Some of the most interesting people are allergic to funnelsThe AutodidactsCuriositryHandy.computer is the Linux dictation app I’ve always wanted.and this post is written with it.The AutodidactsCuriositryA strange workaround for when the capacitive touch screen of your e-reader isn’t responding to dry fingers: use your noseNo, really.The AutodidactsCuriositryAn unexpected benefit of docking my laptop beside my deskI can run Performance profile all the timeThe AutodidactsCuriositrySonder is a word I likeAt a certain point in my life it dawned on me that other people are people too. That guy bent over with Fentanyl? A whole world: boyhood, rites of passage, dreams, regrets, memories, crushes — the works. The evil billionaire? Old wounds, sincere beliefs, time with grandchildren — playing Legos — staring atThe AutodidactsCuriositryTaking responsibility for my energy levelsBootstrapping mitochondrial function one flax seed at a timeThe AutodidactsCuriositryTIL how to accidentally overwrite a decade worth of data in LibreOffice calcAnd also, how to restore from my backup with similar single-keystroke efficiencyThe AutodidactsCuriositryPSA: local AI handwriting recognition is usable nowTo my delight and surprise, local Qwen3-VL:8b WER is acceptable, even on my cryptic handwritingThe AutodidactsCuriositryDe-optimize your search terms to skip the SEO slopFinding the human in a sea of salesThe AutodidactsCuriositryThe Evening Writing EffectTime for some night-owl counter-propagandaThe AutodidactsCuriositrySocial Cache BustingHow do you ask questions that people don’t just hippopotamus?The AutodidactsCuriositryThe Fewer Ingredients The BetterGood quality ingredients hardly need seasoning. I only need to add spices if I’ve burnt the soup.The AutodidactsCuriositryPaste clipboard contents in places where you can’t pasteUse AutoKey send_keys() to put clipboard content places where it can’t normally be putThe AutodidactsCuriositryIt’s surprisingly hard to reliably paste the current date at the cursor on LinuxLong, long ago, in a dotfiles directory far away, I had this snippet: bindsym control+semicolon exec date ‘+%Y-%m-%d’ | tr -d ”\n” | xsel -i -b && xdotool sleep 0.5 key “ctrl+v” Isn’t it a beauty? But it worked, in i3 on X11. I devoted 500The AutodidactsCuriositryWe don’t know until the end, and it’s never the endJudging a book by its content doesn’t work eitherThe AutodidactsCuriositry * * * The questions, corrections, comments, kind words, and good advice we receive by email from strangers and long-time readers means a lot. Thank you! We are also grateful for the donations and paid subscriptions that cover our expenses and keep morale up. We hope to have another 25 new posts for you by the end of march. The Autodidacts

I was perusing my reading spreadsheet, and my reviews of books I read a few years ago don’t always match my current opinion of the book, based on my memory of it. _I was perusing my old book reviews, not because I’m diligent about reviewing the things I’ve learned, but because I was trying to put the pieces back together having recently learned an ominous factoid the hard way: when you have all sheets selected, to search in all sheets,_ [_editing a row in the current sheet overwrites the data in that row of every other sheet_ _.]_ The obvious conclusion is that my recently-celebrated memory is, in fact, feeble, and what I wrote in the spreadsheet cell the day I finished the book is more accurate than what I tell people about the book now. I think this is part of it. I don’t remember most of the details. But there’s another explanation that I like better. **I don’t know whether a book will change my life until it does.** Sometimes, I might have a hunch. Or, I might enjoy reading it. But the effect a book has on me has a lot to do with what sticks with me, and if it changes my thinking and behavior. Some books, which didn’t seem that great, communicated an idea that I took to heart, and acted on. When I think back on those books, I don’t remember the prose, or necessarily even the bad/good parts. I remember, _that book_ _was half-responsible for me becoming a frugality enthusiast,_ or _that book_ _inspired me to become an ultramarathon runner,_ or _that book make me de-clutter my room and give away half my clothing._ Other books, wonderful books, I read too early or too late, or when I had a stomach ache, and they passed through me without leaving a trace. _Or did they?_ If 32,000-year-old seeds buried by Siberian squirrels can germinate and flower, I see no reason to think that impressions planted in my brain a few years ago have expired and become inert. They might be merely dormant, waiting for the right conditions to sprout. Like the story of the old man who lost his horse, it is too soon to say. ****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series.

Long, long ago, in a dotfiles directory far away, I had this snippet: bindsym control+semicolon exec date '+%Y-%m-%d' | tr -d "\n" | xsel -i -b && xdotool sleep 0.5 key "ctrl+v" Isn’t it a beauty? But it worked, in i3 on X11. I devoted 500 words to bemoaning how hard it was to port this over to Wayland + Sway. The section was titled _More Fun with Segmentation Faults,_ and my “solution” involved building alpha-quality software from source, and accepting an 80/20 solution. And then I migrated to a boring version of Linux: Xfce (and later Cinnamon) on Linux Mint. Even though I was on X again, for some reason my original snippet no longer worked 😭. I ended up running AutoKey 24/7 _just so I could have this snippet._ And it didn’t even work reliably. _Further reading:_ _Save the Date_ _._ Paste didn’t work in my terminal emulator, and I spend most of my time in Helix text editor, in Fish Shell, running in a Tmux session, in Ghostty (I know, I insist on making things difficult for myself). I set it to fill the clipboard with the current date so that I could paste it, but I often ended up having something important already _on_ the clipboard when I ran the shortcut, and overwriting it. So today I made a small improvement. It tries to input the date directly with send_keys. But it also _temporarily_ puts the date on the clipboard. Before it puts the date on the clipboard, it backs-up whatever was on the clipboard. Then, after one second has elapsed, it restores the clipboard to its original state. So if I’m in Helix, I hit Ctrl+; Ctrl+Shift+V (which is only as many keystrokes as typing `2026-`, so we are saving ourselves five whole keypresses here! And that’s assuming I know what the date is in the first place, which I usually don’t.) Otherwise, it’s Ctrl+; and done. Here’s my current AutoKey script: import time output = system.exec_command("date -I") # ISO 8601 YYYY-MM-DD format oldtext = clipboard.get_clipboard() # backup what's on the clipboard clipboard.fill_clipboard(output) # put date on clipboard keyboard.send_keys("<ctrl>+V") # paste date time.sleep(1) # give user chance to paste if send_keys failed clipboard.fill_clipboard(oldtext) # restore previous clipboard content It still isn’t perfect, but it’s almost good enough.

****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series. Today I was trying to run an AppImage in a firejail sandbox, to debug the issue that is currently blocking my KOReader Wallabag plugin patch. In the process, I had to paste API keys that were too long to reliably type, and I couldn’t paste into the AppImage with the normal methods. It turns out you _can_ (duh) paste with a long press, just like on an e-reader. But this is so wildly intuitive on desktop that I didn’t think of it at the time. Instead, I wrote a two-line Python AutoKey script, bound to a keybinding (Ctrl+Alt+Shift+Super+k, because all the good keybindings are already taken). cb = clipboard.get_clipboard() keyboard.send_keys(cb) This gets the current clipboard content, and emits it as if it was being typed. Simple, and effective. I’ve run into places where I can type but can’t paste before, so although there was a simpler solution in this particular case, the snippet might be useful in other circumstances.

The day before yesterday I made Borscht. It was delicious, if I do say so myself. Making Borscht is like a conjuring trick: you put a bunch of boring root vegetables in a pot, and it turns into a delicious soup. When I was making this pot of Borscht the other day, I kept being tempted to put some _interesting_ ingredients in it; to spice things up a bit. But the more-experienced cooks I consulted, both in person and on the internet, all advised against it. The secret of Borscht, they said, is its simplicity. A good heuristic for whether a food is healthy is the length of the ingredients list, and the length of the words in that list. If you don’t know what something is, question whether you should be eating it. If a candy bar took that much sweetener and flavouring and preservatives just to make it edible, question whether you should eat it. What are they trying to cover _up_ with all that stuff?! Lousy quality, most likely. Have you ever had plain rice? Plain fresh rolled oats? Straight organic milk? Delicious! Exquisite, just on their own. Sweet. Flavourful. _Addictive_ , even. The same goes for almost every good-quality food I have tasted in its traditional form. But if you have bad-quality, flavourless ingredients, you have to somehow bury that rank taste under a bunch of bold look-at-me spices. It’s like hanging an air freshener in a porta-potty that need cleaning: it fools nobody. And what are those Tocopherols for, anyway? They’re there because the food is so old, or so poorly stored, that it’s actively trying to decay. It has reached the end of its natural life. But somebody wanted to profit, so they pumped it full of preservatives, to make it remain presentable when it really should be feeding the worms. But things that have just a few ingredients don’t give bad quality anywhere to hide. If something’s rotten about it, _you’ll know._ This doesn’t just apply to food. It applies to, like, everything. Websites that have so much bling and hype that you can’t even figure out what product they’re selling. Books that use big words and sweeping statements to create a diversion from the fact that they have absolutely nothing to say that is of any real relevance. There are people like that, too. And songs. The best songs are often very focussed. They say one thing, lyrically, musically, and emotionally, but they say it well. Don’t underestimate a good pot of Borscht.

If you’ve ever tried chatting with a public figure, you probably know what I mean by “hitting the cache”. They produce slick soundbites that sound smart-ish, and could plausibly be connected to the question you asked, or what you said. But the responses aren’t bespoke. It’s like they have a lookup table, and compare the vague topic and sentiment of what you said to their roster of prepared responses, and return the best match. This is not unique to public figures. I do it. I think almost everyone does it to some degree. And the degree tends to correlate with how often they get asked the question. (The same way a webserver serves cached versions of the most frequently-requested, slow-to-load pages.) Since public figures get asked the same questions a lot, it makes sense that they serve most traffic from the cache. The cache can have good stuff in it, but it’s never as interesting as interacting directly with the origin. The cache is stale. The cache is optimized. The cache is _safe_. How do we bust the cache? The first step is to notice that we’re hitting it in the first place. If you are happy with the response you get, there’s no reason to bust the cache. But if you’re talking to a performer, and they have a fake, glassy-eyed smile, and go through all the correct motions, while obviously being totally checked out, _you’re not asking the right questions_. Clearly, you are asking boring questions that everyone asks, and saying boring complements that everyone says: otherwise, they wouldn’t be able to leave the interaction on autopilot. Sometimes, just awareness can bust the cache. The person realizes they’re talking to someone who is _listening_ , rather than just waiting to talk. Respect demands respect. But people giving canned responses sometimes aren’t even there to notice whether the person they’re talking to is paying attention. In this case, a better approach is to try to ask a question that they’ve _never_ been asked before — or make an observation astute enough to pull them back to reality. The question should be something that they are excited to answer: something that makes them learn something new by answering. [1] (Corollary: if you’re an interviewer, you have to do a lot of research, so you can build on everything they’ve already said, instead of starting from scratch.) It sounds like a contradiction that someone could learn something new by answering a question. Isn’t that just spitting out something they already know? No, I don’t think so. Ideas evolve in dialogue, and a good question can demand synthesis of existing knowledge, rather than parroting it; or make someone realize they know something, and have for some time, without knowing that they knew it. In this way, busting the cache is a gift. It lifts the conversation out of the usual fan/boss/co-worker dynamic, into something more alive. It lets the person you are talking to have novel, original thoughts, rather than repeating the thoughts they’ve had before. If someone starts talking the moment the other person finishes, chances are it’s a cache read, and not even a cache read responding to the full content: a cache read that was queued up in response to the _first_ part of what was said. If someone says, “ah, that’s interesting”, or “weird!”, or “hmm, good question”, and then stops to think, agonizingly, for an awkwardly long time, with a sort of bulging wide-eyed look, you might be about to get a fresh shipment of thoughts and feelings. * * * 1. Any question I suggest is, almost by definition, probably not the right one. You can’t demand non-canned answers to a canned question! Besides, I have never managed to feel natural asking questions that I read about in a blogpost. My favourite canned question is, what have you learned recently? I find it doesn’t feel fake to ask this, even if I’ve asked other people that question before. My brother likes to ask, what are your passions in life? Or, what’s new and inspiring in your world right now? The usual “ask open-ended questions about things the person is interested in” seems to apply, but it’s quite different depending on who you’re talking to, and why. ↩︎

****Note:**** this post is part of #100DaysToOffload, a challenge to publish 100 posts in 365 days. These posts are generally shorter and less polished than our normal posts; expect typos and unfiltered thoughts! View more posts in this series. Gwern looked into the morning writing effect, and I read the post with interest. I would like to contribute my n=1 randomly uncontrolled study. This very poor quality, woefully underpowered study concludes: the morning writing effect is fake! The _evening_ writing effect is real. The entire body of research consists of thinking about how many times in the past two months I have gotten up, feeling rested and energetic, and with plenty of time side aside, sat down to write an essay, and discovered that I have nothing to say. And, how many times I have been struck by an idea that feels alive, sometime between 7pm and 2am. It happened this morning. I re-arranged my list of hundreds of essay ideas, sorting them by interestingness. I kept my computer offline. An hour set aside, caffeinated, no excuses. No ideas. Now, it’s 9:30, and I have an early morning tomorrow. Finally I feel like putting words down. I’ve published 46 posts in the past month, and drafted many more, and the pattern has become very familiar. * * * In the comments on _The Morning Writing Effect,_ someone on brought up “the idea that writers lie about their morning writing because it makes them sound virtuous.” But I know that for some people, who aren’t writers, the morning writing effect is real. Some night-owl chronotypes even seem to do their best creative work in the mornings. Maybe I’m just tired? Does that explain my failure of morning writing? No. Even when I go to bed early, get up early, and fatigue isn’t the problem, I feel like my prime time is usually the second half of the day. A number of factors could contribute to _both_ the early morning, and evening, writing effects: * in the morning and evening, we’re less likely to feel like we need to “get to work” on more productive things * in the morning in evening, we’re more likely to be able to work in solitude for extended stretches of time without being interrupted * in the morning and evening, our human, urban, and natural environments are quieter And then there are some factors that are specific to one time or the other: * in the **morning** , we might still be in a groggy, dreamy condition * in the **morning** , our minds might be more blank * in the **evening** , we might have more thoughts and impressions rattling around from the day For me, I think it’s the last one. I started an essay five years ago titled _The_ _Idea Collider,_ touched on the topic in Intentional Computing, and have another one in progress called _It’s scary that the mind is basically a blender._ When I have a long-running project that I work on every day, I can usually get in the flow, if not in the morning, at least by noon. The ideas are partly there, and mainly what’s needed is to load them back into memory, and go from there. It’s harder to start from scratch. I find that when I haven’t been reading, or getting new impressions, I have nothing to say. Perhaps this is the case at multiple scales. (Many things are.) Or perhaps, I’m way off the mark, because it’s late at night and my brain has shut down.

I used to search for “Best Wired Earbuds” and “Best SciFi Films” and so forth, because I like _perfection._ But these kinds of searches don’t get me _the best X,_ they get me _the X with the best marketing team._ These are not the same thing. It took me a long time, and countless affiliate-link-laden “comparisons” before I realized that the top results for “best” anything range from comparing-while-motivated (Wirecutter) to pure unrefined vegetarian snake oil. If there’s money to be had from the people searching for something, somebody, somewhere, has probably spent money to get a particular thing in front of the eyeballs of the populace. As usual, once we understand the system, we can find loopholes in the system. The default thing to search for when looking for opinions about products/media/etc is “best such and such” or maybe “such and such comparison”. But humans often ask questions like, “what’s your favorite such and such?”, or “what X do you use for Y?”. There’s a Venn diagram of what people search for, and what people SEO optimize for: This human crescent on the right is where we can find peace, and good quality information. Now excuse me while I go search for the second best wired earbuds, fairly decent scifi movies, and serviceable camping tents. Perhaps we will meet again in the comments of one of those “less-well-known generalist blogs.”