Posts

Did you know that the transformer—the architecture underlying nearly every modern AI model, including physical AI—was originally developed for machine translation 1 ? It’s been almost five years since I launched this website, and I’ve written more than 100 posts in that time. But my friends abroad probably don’t know that, because unlike the rest of the website, the blog has only been in Korean. I haven’t had time to translate every post as I write it. So ever since AI’s main talent was translation, years before it became nearly capable of everything, I’ve wanted to exploit it (?) to build a pipeline that would translate my posts automatically. I couldn’t even find time to build that, so I put it off for more than two years. But after spending a little time and a few tokens, I’ve finally added it to the website as a new feature. Coding agents really have gotten good—I built it in no time. Now, if you set the site language to English, you’ll see translated posts instead of the Korean originals. I think it’ll be a different experience from simply running the originals through Google Translate or DeepL. The translations follow instructions I’ve tailored to my preferences, and I review them afterward. More importantly, there’s a big difference between a post being available in English and having to press a translate button to read it. That’s probably why X (formerly Twitter) recently made the big decision to translate all posts and replies proactively. This will be the first post translated that way. Earlier posts will be translated soon, too. To all my friends around the world: let’s use the wonders of modern technology to feel a little closer. 1 More precisely, sequence transduction, which includes machine translation.

I had an errand in Sejong this evening, so I made a quick trip there. After spending day after day at my desk trying to finish my dissertation, the outing cleared my head. An interesting thought came to me in the car, so I’m writing it down. The title means that, as of today, I’ve been alive for 27 years, 2 months, and 7 days. I didn’t plan it, but those numbers look rather nice. Perhaps they’re telling me to write this today. I want to write about what’s happening at this point in my life. I’ve written before about things I’ve come to see differently as I’ve grown older. Those pieces were about changes that depended on me—on what I encountered, read, and thought about. This one is a little different. Today I want to talk about what I’ve observed as the consensus among people my age at this particular moment. Since I’m describing a “representative value” that averages out individual differences, my exact age seemed like a fitting title. It might be fun to write a life-version update log like this from time to time. I call it a generational consensus, but my observable universe is very small, so I should say up front that it may differ greatly from the ground truth. If anything, this post might show just how narrow the world I experience is. v27.2.7 Update Many people my age now live in households of their own—unless they live and work in the same area as their families and the city’s punishing housing prices make moving out prohibitively expensive. We sometimes visit one another’s homes. There’s much less need to fret over finding a café or bar that stays open late. Some have cars, and even those who don’t can sometimes afford a taxi, so the last bus or train is less of a concern, too. Not many are married. Women tend to marry earlier than men, so a few cases are starting to appear, but there still aren’t many. I suspect that will change substantially before long—perhaps within the next year or two. Someone I met recently, only a few years older than me, said that about half of his friends seemed to be married. I was honestly a little surprised, especially since he was a man. Even with the age of marriage rising, the wave is coming. A wave of childbirth and parenting will soon follow. Those who aren’t the oldest child can often already watch an older sibling ride that wave up close. As someone who prefers quiet places, I felt I was at the peak of independence in my life while sitting at home with friends, chatting into the late evening. But if we intend to marry—myself included—this period won’t last long. It shouldn’t. It’s a brief golden age in which we have the means to live independently but don’t yet have families of our own to care for. To start a family, we must be ready to give up that golden age and accept the merging of our worlds with those of a spouse and children. Back in high school and even early in college, differences in aptitude mostly prompted thoughts like, “Oh, this friend is good at that.” But a substantial amount of time and effort has since built up on those differences, and we’ve now branched into entirely different species species according to our fields. Our ways of life have diverged that much, and staying in touch with people of another species takes a special effort. Few people are just starting out in the arts or sports now. Quite a few are closer to the end of their careers than the beginning. Many are entering their prime or are already in it; in certain fields, such as professional Go or esports, that prime has already passed. Success tends to come later for people in academia, the professions, and the startup scene—including me and many people around me—than it does for those in the arts or sports. Almost no one in my circle has achieved definitive success yet, and even looking beyond it, I could count such people on one hand. But I’m starting to see people who have a good chance of getting there. I look forward to seeing, in five or ten years, how sound my judgment was. The moment for people in business and politics comes later still. Even beyond my own circle, it’s hard to find anyone my age who has so much as entered those worlds. The end of life no longer feels vaguely, immeasurably far away. Our parents have already retired or soon will, and with each passing day it becomes clearer that we should be caring for them rather than receiving their care. Many of the public figures whose deaths are announced are people whose names I already know.

--> Lately, I’ve been using Codex to help modify codebases for my research, and I keep running into its usage limits. The way those limits work is interesting: Usage is measured at two levels: five hours and one week (I understand Claude Code has a similar system). There are two straightforward ways to define the interval used to determine whether a limit has been exceeded. Taking the 5h level as an example: Fixed intervals, such as 9 a.m.–2 p.m., 2 p.m.–7 p.m., and so on (fixed window) or The 5h immediately preceding a usage request (moving window) Codex works somewhere between 1 and 2. If a user stops using it for a while and then resumes, a 5h window starts when they resume and resets five hours later. If someone used Codex around the clock, this would work much like option 1, but that’s not realistically possible. This odd way of measuring usage creates a paradox: by using Codex more (strategically), you can actually increase how much you get to use it. Here’s a simple example: Suppose someone arrives at work, does something that doesn’t require Codex, then uses up 100% of their 5h allowance on a Codex task at 11 a.m. Unless they buy additional credits, they can’t use Codex again until 4 p.m. But if they had done a quick task with Codex as soon as they arrived at 9 a.m., they could do another task at 11 a.m. that uses nearly 100% of the allowance, and their usage would reset at 2 p.m., giving them another full allowance. You can exploit this to stretch your usage as far as possible. The key is to schedule a cron job that has Codex perform a task at set intervals starting early in the morning. That way, the reset happens automatically in the middle of the hours when I expect to be working most intensely. For me, 10 a.m., 3 p.m., and 8 p.m. would be good. The lighter the task, the better. I simply set it to shout “Codex, woohoo-!”: This fixes the daily reset times at hours divisible by five without using even 1% of the 5h allowance. In conclusion: Codex, woohoo-! *** Update #1 Someone saw this post in my Instagram story and tipped me off: more than six months ago, someone had exactly the same idea and posted a GitHub Gist with a cron job script and an explanation of why they developed it. It also includes scheduling for Claude Code, so take a look if you’re interested. Update #2 The cron job in the main post is just a baseline meant to illustrate the idea; you can add a few more features. Here’s the job I actually set up on my computer: ~/bin/codex-log-test.sh #!/bin/bash LOGFILE="$HOME/logs/codex.log" { echo echo "==================================================" date '+%Y-%m-%d %H:%M:%S' echo "==================================================" codex exec \ '"Codex 야호-!"라고 출력해.' \ --config model_reasoning_effort="low" \ --skip-git-repo-check } >> "$LOGFILE" 2>&1 terminal chmod +x /Users/gunheeyi/bin/codex-log-test.sh $ crontab -l 56 4 * * * /Users/gunheeyi/bin/codex-log-test.sh 58 9 * * * /Users/gunheeyi/bin/codex-log-test.sh 0 15 * * * /Users/gunheeyi/bin/codex-log-test.sh 2 20 * * * /Users/gunheeyi/bin/codex-log-test.sh This adds the following: exec mode & the --skip-git-repo-check flag: To run reliably as a cron job, it needs to run in non-interactive mode and skip the directory check. >> "$LOGFILE" 2>&1 : Saves the output to a log file so I can check whether cron ran successfully. echo ~ : Records the date and time in the log alongside the output. --config model_reasoning_effort : Sets reasoning effort to its lowest level to minimize token usage. The jobs run 5 hours and 2 minutes apart, rather than exactly five hours apart, in case a job runs faster than the previous one and therefore falls within the previous window without resetting it. (I haven’t tested this.) One minor caveat: cron is configured on the local computer, so it won’t run while that computer is off. If you want the job to run reliably, you should leave the computer on or, ideally, schedule it on a server if you have one. Usage is shared among sessions on the same account rather than measured per device, so opening a window on a server should make that same window apply on other computers too. I haven’t rigorously verified this yet.

Reading this now. Borrowed it along with “Things I Wish I’d Known as a Graduate Student”—the start of an epic late-to-the-party reading series. Why are there so many things I should’ve known?

If the book that started last year was Han Kang’s Human Acts , this is the one that opened 2026. It was a good choice. After twenty-some years of life, I feel as though I’m only now beginning to understand how to live. Your words, written after eighty-some years of life, reassure me that what I’m learning isn’t wrong. Thank you.

Not long ago, for my father’s birthday, I gave my parents new phones. A gift doesn’t have to be something the recipient needs, but it should be something they’ll be happy to receive. As for whether my parents needed new phones, they’d needed them for quite a while. To me, both phones had looked like they were on their last legs for ages, but my parents insisted they worked just fine… Anyway, they were delighted with the gifts. My father’s birthday falls near the end of the year, close to Christmas. That got me thinking: this year marks exactly 20 years since I received the Christmas present I remember most vividly. At Christmas in 2005, Santa Claus gave me an electronic dictionary. It was the first electronic device I ever owned. It had a monochrome reflective LCD with a refresh rate of maybe 3 Hz, and hardly any features beyond a dictionary, notes, a calendar, and contacts. But I was probably even more excited about it than I was when I got my first MacBook. I remember spending days exploring every menu, writing diary entries in the notes app, and saving my friends’ home phone numbers. I used a feature phone until I was an adult, and I didn’t have a personal computer until high school. So I kept this dictionary close for at least ten years. I think I even left it in the study hall to use when I was in high school. I learned so many words from it. Of the words I used in a seminar I gave in English last week , I probably learned more than half from that dictionary. Looking at it now, it’s such a tiny, cute electronic dictionary. I put it through so much that I had both the top and bottom halves of its case replaced once. Truly the electronic dictionary of Theseus. That Christmas had an unusual setting, too. My uncle’s housewarming happened to be scheduled for that day, so I received the present at his place. Little me could only marvel: “How did Santa know where my uncle lives??” Later, my parents Santa Claus told me that coordinating plans with the relatives kept changing the location, and figuring out how to transport the gift along with it had taken some work. The funny thing is, I faced a similar dilemma this year. As I talked with my younger sibling about whether to risk having the phones delivered directly to my parents’ home or have them sent to mine and bring them up myself, I realized my parents must have had a similar conversation in 2005. Exactly 20 years later, our roles had completely reversed. Of course, I still rely on my parents in some ways, but it was a moment when I realized I’d grown up enough to give them gifts, at least now and then. And isn’t it fitting that the electronic dictionary and the smartphone could each be seen as a defining electronic device of its time? The difference in what they can do made me feel just how much time had passed. There’s one more happy memory I want to share. At the end of 2003, my family went to Bangkok, so I received my Christmas present during the trip. That year’s gift was a magic drawing board—a fairly fancy one, with a lever on the side that, when pulled down, would slowly spring back up and erase the whole drawing surface at once. I got a lot of use out of it, drawing whatever I wanted and writing things down as I learned them. But what struck me most that day wasn’t the gift. It was the letter that came with it. The letter was a whole page in Thai! It had never occurred to me that Santa might be from another country, so I was dumbfounded—in a good way. I probably went to my parents to tell them all about this strange and wonderful thing. I must have been quite astonished, because I still remember bits of it even though I was only five. My parents had actually written the letter in English and asked a hotel employee to translate it into Thai. Receiving that letter and gift remains one of my most treasured memories. Someday, I want to do something like that for my own children—or come up with something even more imaginative. The funny thing about gifts is that you put all the care into giving them, but later, the ones you remember are the gifts you received. It’s not as if I’ve never given anyone a gift, yet the ones that come to mind are all things people gave me. When I’m the one giving, I worry about whether the recipient will like what I’ve prepared. But when I’m receiving, it doesn’t really matter what it is. I’m simply grateful that someone thought of me and picked something out. Recently, someone gave me a plush toy. I’d never been interested in plush toys before, but I liked it so much that I’ve put it on the back of my sofa at home. A gift of something unfamiliar can also become a chance to discover something new. To my elementary school friends who gave me a Pentel mechanical pencil for my birthday in 2012: did you know I still use it? I liked it so much that, even though it’s broken a few times, I’ve gone out of my way to buy the same one two or three more times. I even used it for a quantum computing exam recently. Maybe because KakaoTalk makes it so easy to see and keep track of birthdays these days, gifts that arrive unexpectedly seem to mean more to me than ones given on occasions like birthdays. To return to that plush toy: a senior colleague in my lab bought it while attending a conference abroad. I was supposed to go too. I’d submitted an abstract, had it accepted, and even finished writing the paper; all that was left was to fly out. But something absurd came up, and I couldn’t go. It was the first international conference I’d gotten into since starting my degree, and my colleague thought of me, unable to attend at the last minute, and brought back the toy. I was so grateful for the thought. The aerospace conference was in Sydney, so it’s a kangaroo in a spacesuit. In a world where everyone is busy just managing their own lives, remembering someone when it isn’t even their birthday takes thoughtfulness and effort. As the year draws to a close, thinking about the people like that in my life warms my heart all over again. I want to try harder to be that kind of person, too.

13th Global Trajectory Optimization Competition Work in progress

I once wrote a post called The Number Devil about how we continue to grow mentally even after we stop growing physically. That was in May 2022, about four years after I became an adult. Roughly the same amount of time has passed again. I think I’m still growing. There are things I can do now that I couldn’t do then. Some are small, like getting my gloves out and putting them on before the cold makes me miserable (amazingly, I think this is the first year I’ve managed to do that properly). Others are fundamental enough to help me in every part of my life. Today I want to write about one of those. Why do humans have emotions? Having emotions helps us survive. Fear makes us flee danger, and anger makes us fight threats. Hope motivates us to work toward long-term goals, while love and empathy play important roles in forming social bonds and living together. But emotions don’t always work in our favor. Sometimes they ruin things. When I’m overjoyed, that’s not really a problem. It means something wonderful has happened, so who cares if a few other things go wrong? Joy that intense is rare and short-lived. I might as well enjoy it. The opposite is the problem: when negative emotions take over. When something bad has already happened, and the feelings it leaves behind threaten to ruin other things in a chain reaction. At times like that, I’ve wished I could briefly become a robot—one that feels nothing and acts only on reason. As if that were possible. If you’re human, you can’t help having feelings. What matters is how you respond to them. Over the years, I’ve developed a manual for dealing with mine: *** First, consider how long it’s been. How long have you been feeling this way? 1. If the feeling has just appeared, it’s easy. Emotions that arise in an instant fade quickly, too. Anger, for example, is said to take only about 90 seconds for the chemicals that trigger it to be produced and then dissipate. So you don’t need much of a strategy. You just have to get through that one- or two-minute window. For that brief time, refrain from saying or doing anything driven by the emotion. Park Yong-maan, former chairman of Doosan Group, said in a documentary that he keeps two- and five-minute hourglasses on his desk. Before acting on an important decision, he turns one over and waits for the sand to run out. By then, he said, the emotion has drained away, and many of his decisions change. That stuck with me, so I bought a three-minute toothbrushing hourglass at Daiso and put it on my desk. Fortunately, I haven’t yet had a moment heated enough to make me turn it over. 2. Emotions that have been building for a long time, or that linger, need a more careful approach. 2.1. First, find the cause. Why do I feel bad right now? Once I’ve identified it, I ask whether it warrants feeling this strongly. Questioning whether my own feelings are justified might sound a little psychopathic, but it genuinely helps. Surprisingly often, the cause turns out not to be such a big deal—or there’s no particular reason at all. For me, the former has usually been disappointment in someone, and the latter, a low mood. Feelings I’ve found to be unwarranted at this stage are relatively easy to shake off. 2.2. The next step for feelings that do have a reason behind them is to change my perspective. I can’t undo what happened, but I can change how I take it. In psychology, this is called reframing. Suppose I completely botch a presentation at a conference. Rather than see it as a failure and wallow in frustration, I can see it as a chance to address my weaknesses and do better next time. That not only makes me feel better in the moment; it also leads me to analyze and document my mistakes from different angles and prepare more systematically for the future. reframing 2.3. Still not enough? Then fight back. Identify a few active things you can do to overcome the feeling, and do them whenever you feel bad. These aren’t quite the same as things that make you happy, like hobbies. Those can lift your mood in one area and offset the bad feeling in another, which certainly helps to a degree. But the activities I mean can bring the bad mood itself back toward neutral. What works will be different for everyone. Here are a few things that have worked for me: Sleep It has a good chance of dispelling the thoughts that come late at night. So if it’s late, stop thinking about everything else and get some sleep. Aerobic exercise Something you can do at a steady pace for a long time is good, especially if you can look at the scenery while you do it. Running outdoors and cycling, for example. Personally, anaerobic exercise hasn’t helped me much with regulating my emotions. Watching a movie Playing an instrument As I write this, I’ve just realized that all these activities have something in common. Sleep is obvious. During aerobic exercise, I have to manage my breathing and pace; when watching a movie, I’m absorbed in the protagonist’s situation; and when playing an instrument, I’m trying not to miss a single note. Each demands my full attention, and I have to stay focused for at least several dozen minutes. That forces me to spend some time apart from whatever hurt me. Perhaps that’s exactly why these activities help me recover. 2.4. But what if an emotion still won’t come under control after all those steps? Then there’s nothing left but the last resort. Think selfishly. Is staying like this doing me any good? Sometimes, no matter how I look at it, the world really does seem to be treating me unfairly, and I can’t find a single positive angle. Some experiences leave feelings that a good night’s sleep and a run won’t resolve. But am I going to spend the rest of my life frustrated and cynical because of them? That won’t do me any good. When a situation is already bleak, the only way to make even a small positive change is to get moving. The person who breaks free of an emotion’s grip and acts is the one who ultimately wins. I added this final trick to my manual after watching a YouTube video recommended to me one day. I hope you’ll watch the video I’ve embedded below. Can a one-minute video change your life? I don’t usually believe that sort of thing. But this one changed mine. Not completely, but it made me more resilient, better able to overcome the many hardships I’ll face in the long years ahead. The full version is here . When repeated effort brings me nothing but failure, and I feel myself growing cynical, I always think of Conan’s words. *** I think back to myself just a few years ago, before I had this manual. I was so young. I’d sulk over little things, or sink into a low mood for no particular reason and get pulled deeper and de

A person dies and leaves a name behind. I’ve been thinking about this saying a lot lately. I want to leave my name behind. I want those who knew me personally to remember me as a person of good character, and those who didn’t to remember me as an engineer, entrepreneur, and creator who made the world a better place. At the same time, I realize how hard it is to leave a name behind. Of the more than a hundred billion people who have walked this earth, how many can you name? Not long ago, I became captivated by the songs of a singer who had died young and listened to nothing else for days. The singer had never married and had no children. From a biological perspective, an organism’s greatest success is to leave as many copies of its genes, or something like them, in the world as possible. Conversely, an organism that leaves no offspring has failed. By that measure, the singer was a failure. But was he really? Even now, long after his death, I listen to his songs. His creations have outlived him and continue to be enjoyed by countless people. His music comforts us and gives us the strength to face another day. Helping people who share my values live happier lives—that is what I mean by “leaving my name behind.” From a broader perspective, it is also a way of passing on my genes. It may not be as exact a copy as biological reproduction, but it can reach far more people and endure indefinitely. Reproduction is now just one of many ways to pass on genes. We can see evidence of this in how modern nations are changing. The ultimate form of a nation is a community of people who share the same values, not the same race. The boundaries between nations once drawn along racial lines are blurring with globalization; now ideas and ideologies bring people together. In God We Trust: because of its unique history, the United States was the first country to approach this ultimate form. Bonds forged through shared values can sometimes be stronger than bonds of blood. We often see people join forces with someone of a different race who shares their ideology, then stand against—and even point guns at—people of their own race. Evolutionary biology, with the preservation of genes as its first principle, struggles to explain such behavior. Or perhaps one day, when genetic research has advanced far enough, we’ll discover that the human values we thought too abstract to be encoded in a sequence of bases are, in fact, inscribed in broad patterns across our DNA. If so, the flourishing of a group that shares the ideals I support could truly count as a biological success, just as much as having sons and daughters of my own.

The file explorer will load here. If it doesn't appear after a while, try refreshing the page. When I was young, I would sometimes see professors being interviewed in clips on TV. Often, I wondered whether the producers really needed to ask a professor for answers like “As the weather gets colder, drink plenty of warm water to prevent respiratory illnesses” on the news, or “Yes, it's true. When the novel <Jaws> was first introduced in Korea, it came out under the title <Agari>” on Sponge . But I suppose they wanted to show not just the answer, but also the credibility lent to it by the professor's position. The bookshelves full of books behind the professors made them seem all the more credible. Looking at those shelves, I vaguely wondered, “Will I have a bookshelf like that when I grow up?” There they are, doing just that About twenty years later, I have grown up, but I still don't have a bookshelf like that. A few books and a few stacks of papers I've printed myself are all the materials at my desk in the lab. My sparse workspace isn't just a reflection of how short my research career has been. Even my advisor, who has spent fifteen years at our university, keeps relatively few books on the shelves compared with the professors I used to see on TV. My desk in the lab today (November 30, 2025) I think the digitization of materials has played an even bigger part in simplifying our workspaces. Today, we handle most of our materials on computers. As with this very post, I spend far more time reading and writing with a screen and keyboard than with paper and pen. Chances are you do too. As the amount of material on my drive grew, I needed a system for managing it sustainably. I began developing one around high school, when I got my own computer, and have kept refining it ever since. It has now settled into a fairly stable form. I imagine nearly everyone who does knowledge work for a living has a system of their own. Yet we rarely get to look inside one another's systems. Physical bookshelves could be cumbersome, but they also offered evidence of our past efforts—the books behind those professors—and chances to learn from someone else's work by accident: “Oh, what's that book?” Digitizing our bookshelves has taken much of that charm away. So I decided to write a post actively introducing my digital bookshelf. The file explorer at the top lets you explore the actual structure of my archive, and what follows explains some of its more noteworthy parts. Text like this refers to the name or path of a particular directory or file in the archive. ~ is shorthand for my home directory ( Macintosh HD/Users/gunheeyi ) . My YouTube recommendations one night while I was writing this post. Coincidence? Directories Here are the main directories (folders) that make up my archive. ~/iCloud Drive With a few exceptions, I store everything in iCloud. I currently use four computers, switching between them depending on where I am and what I'm doing. Cloud storage is essential if I want to work seamlessly across all of them, as well as my tablet and phone. Back when I used a Windows PC and a Galaxy phone, Google Drive served me well. But after moving into the Apple ecosystem, I moved all my materials and photos to iCloud. What I like most is that it is built into the operating system on both desktop and mobile, so it works well without any separate installation. The first exception to cloud syncing is files downloaded from the web. They go into ~/Downloads . After checking their contents, I move those worth keeping long-term into a directory under iCloud. Otherwise, I delete them immediately or during a periodic cleanup of the Downloads folder. The other exception is the material in the dev directory, which I'll explain later. ~/iCloud Drive/yearly yearly is the largest directory in my iCloud Drive. It holds almost everything that can be organized chronologically. As the name suggests, its immediate subdirectories are years such as 2025 . Within each year, I sort materials by subject. Under 2024 , for example, are 선보엔젤파트너스 (the company I worked for before graduate school), kaist (research and coursework), founders (a student entrepreneurship group I was active in early that year), freelancing (contract development work), and misc (miscellaneous odds and ends). Of course, I could have chosen subjects rather than years as the top-level categories. But thinking in terms of when and then what feels more natural to me. I also like being able to see at a glance what I was or am doing in a given year, so I chose to organize by year. Sometimes work on one subject crosses into a new year. If it doesn't continue far into that next year, I put everything under the first, primary year. That applies to an academic year running from March through the following February, for instance, or a development contract that began one summer and ended the following January. When material spans multiple years substantially, I inevitably end up with directories of the same name under several years. ~/iCloud Drive/yearly/future One of my most recent changes was adding a future directory under yearly . It holds material about things I expect to happen in the future but whose timing is still uncertain. If the timing is known, the material goes directly under a specific year instead. Examples include marriage, raising children, postdoctoral research, finding a job, and trips I haven't taken yet. Since these things haven't happened, the contents are mostly wish lists, tips that may come in handy when the time comes, and information about support programs rather than records of the past. Now that future exists, would chronological be a more accurate name than yearly ? ~/iCloud Drive/resources Things I need to keep but that aren't tied to a particular time go in resources : books, favorite works of art, sheet music, contacts, software, operating-system and application settings, wallpaper images, fonts, document templates, and more. ~/iCloud Drive/tunnel I use tunnel to move material between devices. Other directories in the cloud hold things up in the “cloud” so I can download them whenever I need them. tunnel , as its name suggests, is more like a