flâneur

Feranno Liu

1 followers · 8 following · 343 views

on the atlas — 39

highlights — 59

  • During the trip, Yang finished about 40 books, mostly non-fiction, history, and travelogues. Whenever he sent a letter to his family, he’d include the books he’d finished reading
    134 Days, 68 Places, Zero Internet: One Man’s Journey Through Digital China
  • Unable to find an ATM or a shop that accepted physical cards, he went from store to store, asking shopkeepers if they could give him cash using his bank card. His unusual request eventually prompted one skeptical local to ask if he was laundering money.
    134 Days, 68 Places, Zero Internet: One Man’s Journey Through Digital China
  • At Chifeng bus station in the northern Inner Mongolia Autonomous Region, he tried to buy a bus ticket 100 kilometers south to Ningcheng County, but couldn’t pin down exactly which town he wanted to go to — hard to do without a quick search. A station worker accused him of being a spy because “only spies don’t use phones, since they are afraid of being tracked.”
    134 Days, 68 Places, Zero Internet: One Man’s Journey Through Digital China
  • Yang visited some big or mid-sized cities at first, but later felt they shared the same vibe, so he turned to smaller cities and rural areas instead, hoping for “more authentic experiences and diverse landscapes.”
    134 Days, 68 Places, Zero Internet: One Man’s Journey Through Digital China
  • Another issue was departure times, which Yang could not easily check without a phone. Whenever he wanted to leave a place, he would pack his things, head straight to the train or bus station, and simply buy a ticket for the next available departure.
    134 Days, 68 Places, Zero Internet: One Man’s Journey Through Digital China
  • At his first stop, a chain hotel in Linfen, another city in Shanxi, he couldn’t get a room in person because the hotel only accepted online bookings. The front desk staffer told Yang there was nothing he could do to help him check in, though kindly drew him a map to another location of the same chain that could accommodate walk-in guests.
    134 Days, 68 Places, Zero Internet: One Man’s Journey Through Digital China
  • Into Yang’s backpack went several books, two cameras, his ID card, a few changes of clothes, two paper maps, some cash and bank cards, a notebook and a pen, and — in a deliberate nod to the analog world he was about to reenter — several Chinese writing brushes, ink, and xuan paper for writing letters. No phone. No data.
    134 Days, 68 Places, Zero Internet: One Man’s Journey Through Digital China
  • Lexical tokenization is the conversion of a raw text into (semantically or syntactically) meaningful lexical tokens, belonging to categories defined by a "lexer" program, such as identifiers, operators, grouping symbols, and data types. The resulting tokens are then passed on to some other form of processing. The process can be considered a sub-task of parsing input.
    Lexical analysis
  • Even in his early years, Linnaeus seemed to have a liking for plants, flowers in particular. Whenever he was upset, he was given a flower, which immediately calmed him. Nils spent much time in his garden and often showed flowers to Linnaeus and told him their names. Soon Linnaeus was given his own patch of land where he could grow plants.
    Carl_Linnaeus
  • During Linnaeus' lifetime, Systema Naturae was under continuous revision. Progress was incorporated into new and ever-expanding editions; for example, in his 1st edition (1735), whales and manatees were originally classified as species of fish (as was thought to be the case then). In the 10th edition, they were both moved into the mammal class.
    10th edition of Systema Naturae
  • a book written by Swedish naturalist Carl Linnaeus and published in two volumes in 1758 and 1759, which marks the starting point of zoological nomenclature. In it, Linnaeus introduced binomial nomenclature for animals, something he had already done for plants in his 1753 publication of Species Plantarum.
    10th edition of Systema Naturae
  • A context-free grammar provides a simple and mathematically precise mechanism for describing the methods by which phrases in some natural language are built from smaller blocks, capturing the "block structure" of sentences in a natural way. Its simplicity makes the formalism amenable to rigorous mathematical study. Important features of natural language syntax such as agreement and reference are not part of the context-free grammar, but the basic recursive structure of sentences, the way in which clauses nest inside other clauses, and the way in which lists of adjectives and adverbs are swallo…
    Context-free_grammar
  • My political ideal is democracy. Let every man be respected as an individual and no man idolized. It is an irony of fate that I myself have been the recipient of excessive admiration and reverence from my fellow-beings, through no fault, and no merit, of my own. The cause of this may well be the desire, unattainable for many, to understand the few ideas to which I have with my feeble powers attained through ceaseless struggle. I am quite aware that for any organization to reach its goals, one man must do the thinking and directing and generally bear the responsibility. But the led must not be …
    "The World As I See It" by Albert Einstein
  • A hundred times every day I remind myself that my inner and outer life are based on the labors of other men, living and dead, and that I must exert myself in order to give in the same measure as I have received and am still receiving...
    "The World As I See It" by Albert Einstein
  • Scaling laws are one of the most critical empirical findings in deep learning. The observation is simple in form: the training loss decreases predictably as we scale up model size , dataset size , and compute , following a power-law curve, which appears as a straight line on a log-log plot. We can view scaling laws as a framework for describing the relationship between compute, loss, model size and data; at its core, it is about how to allocate precious compute optimally between and .
    Scaling Laws, Carefully | Lil'Log
  • Unless your income is really quite modest, you can afford to buy and regularly view some pretty high-quality Indian textiles. In India I’ve found some excellent pieces for as cheap as $200-$250.
    How to understand modern India - Marginal REVOLUTION
  • Studying textiles and cloth also brings you right to Gandhi’s “Swadeshi’ movement.
    How to understand modern India - Marginal REVOLUTION
  • People who write books on textiles tend to be both clear and careful I have found, perhaps because they love and collect something delicate.
    How to understand modern India - Marginal REVOLUTION
  • Books on textiles will explain the accompanying information about Indian history in a clearer way than will actual history books about India.
    How to understand modern India - Marginal REVOLUTION
  • In the early 18th century, India was a world leader at cloth production, but it lost this position by the early 19th century. Studying textiles and cloth production offers an excellent window on their major story of economic decline, and how British import penetration, backed by colonialism, contributed to Indian deindustrialization.
    How to understand modern India - Marginal REVOLUTION
  • It focuses your attention rather immediately on the role of women and women’s work, and also how this interacted with industrialization.
    How to understand modern India - Marginal REVOLUTION
  • Concentrating on the algorithm for a basic arithmetic operation was typical of Richard's approach. He loved the details. In studying the router, he paid attention to the action of each individual gate and in writing a program he insisted on understanding the implementation of every instruction. He distrusted abstractions that could not be directly related to the facts. When several years later I wrote a general interest article on the Connection Machine for [Scientific American], he was disappointed that it left out too many details. He asked, "How is anyone supposed to know that this isn't ju…
    Feynman’s Rigor « the jsomers.net blog
  • The bureaucracy itself might become increasingly opaque to human oversight.
    Gradual Disempowerment
  • Democratic processes might persist formally but become less meaningful.
    Gradual Disempowerment
  • The complexity of AI-driven governance might make it increasingly difficult for human citizens to understand or critique government decisions.
    Gradual Disempowerment
  • A state where AI systems have replaced human labor in many facets of governance — such as administration, security, and justice — could provide some enormous boons. On the surface, it might appear highly efficient and even benevolent. We might see lower crime rates, less low-level corruption, greater tax revenues, and more efficient public services. At the same time, the gradual replacement of human involvement in governance could lead to a subtle but profound shift in the relationship between citizens and the state. Even if the system appears to function well, citizens might find themselves i…
    Gradual Disempowerment
  • To do this in real-time, you need to practice breaking down arguments. Now say you’re approached with a topic out of complete left field—”Monkeys in Narnia are instituting mandatory curfews”. Here the argument might seem absurd and pretty different at first glance from consuming data, but it can be broken down into smaller building blocks! Mandatory curfews sound a lot like government intervention, curfews might relate to the literal concept of curfews in highschool/parent psychology, etc. Then we can cross apply the pro/con arguments relating to these building blocks to the monkey situation.
    how to think fast
  • By having more granular (again, not necessarily more data) building blocks, you can get a better hang of new concepts that are deep out of your comfort zone.
    how to think fast
  • When people ask me what I’ll do when there’s nothing left that I’m comparatively good at, I always tell them something like gardening. Almost no one grows tomatoes because they can do it better than the agrico-industrial complex; it’s more scale model than economy of scale. But it’s fun to see a thing that you put into the ground turn into something you can eat, isn’t it? Fun and oddly personal. And almost every gardener I know swears their produce tastes better, much better...
    Biases, Weights - Aidan Smith
  • whenever there is a tension, tradeoff, or decision, have conviction that there is a right answer
    Being a Force of Nature
  • I do think quite strongly that the next interface looks something like this – a generalist agent that helps you out. In the limit, I think this gives very high valence futures, and it’s a future that I’m excited about. It gives you the benefits of modern technology – knowing what your friends are doing, being able to talk to your loved ones, being able to know the odds on events (and maybe even gamble a bit tee hee) – while giving you the aspects of humanity that you value. It gets your ass out the door instead of keeping you in your room all day with social media feeds, for example. In the fa…
    Cautious Technooptimism | Sudarsh Kunnavakkam
  • There’s a sentiment going around that this way of going about technology is a bad way to derive valence, that we’re giving up some form of long form happiness by focusing on joy and productivity; that giving up on tech and monkeying out is the right way to get find fullfilment, restricting your usage of technology to your work where you get more productive. I question this view heavily! The ability to video call a loved one across the Atlantic is meaningful, being able to fly anywhere in the world for a low price allows you to broaden yourself, so on and so forth. Even in the restriction to so…
    Cautious Technooptimism | Sudarsh Kunnavakkam
  • the easiest way to tell is to simply increase the size of your data. If that doesn't increase the runtime proportionally, you're overhead bound. For example, if you double your batch size but your runtime only increases by 10%, you're likely overhead bound.
    Making Deep Learning Go Faster
  • frameworks like PyTorch also have many layers of dispatch before you get to your actual kernel
    Making Deep Learning Go Faster
  • So why do our non-matmul ops take so much more time than they should?
    Making Deep Learning Go Faster
  • So, you want to improve the performance of your deep learning model. How might you approach such a task?
    Making Deep Learning Go Faster
  • The core realization with FlashAttention was that the size of the intermediate computations dwarfed the actual size of the inputs and outputs. I.e., the 𝑄 𝐾 𝑇 QK T matrix has size 𝑂 ( 𝐿 2 ) O(L 2 ), even though the inputs themselves are only 𝑂 ( 𝐿 ) O(L)
    Mamba No. 5 (A Little Bit Of…) | Sparse Notes
  • Speculatively, RNNs and transformers have a limited lifespan because they make poor use of increasingly abundant compute. It’s critical that we design models that better leverage compute while also maintaining or preserving fundamental model quality. Hence, the interest in Mamba.
    Mamba No. 5 (A Little Bit Of…) | Sparse Notes
  • We should reduce the cognitive load in our projects as much as possible.
    Cognitive load is what matters
  • I have very mixed emotions. There were high highs (well, I guess the competition day) and lots of low lows. :( I guess it’s a big and exciting project, but at the same time as a Caltech student I felt like it was just some random side-quest that I was sinking way too much time in. People around me were getting real internships and doing real research, and I was just sinking all my time into this random… high school math competition? Surely that could not be the right thing to do. I was head of both Problem Writing and Logistics (first time being head of either) because no one else really wante…
    Solitude and Society - by Vivian Loh - circles
  • On January 13th, I frantically made a document titled “CMM T-” in which for each remaining day I listed out things that should ideally be done on/by that day. On January 14th, I went to an administrative office to figure out how much money was in our bank account (because we still had quite a few things to pay for) and realized that we were a little short on money so I panicked and reached out to sponsors again last-minute, and fortunately we ended up fine in the end. On January 15th, I ordered 480 bagels and 600 subway sandwiches with the help of an administrative official (because I didn’t h…
    Solitude and Society - by Vivian Loh - circles
  • During the last two weeks between January 10th and January 24th, I became exponentially more cooked. My classes also ramped up during those two weeks — winter term, which is 10 weeks long, started on January 5th, so fortunately the CMM grind period was a the beginning of term when I tended to be less busy, but I was still attending classes every day and by the 2nd week I started having homework that was quite nontrivial. Right after the Power Round was released, I answered all the questions that were pouring in about it and fixed many mistakes in it. During the 2nd-to-last week (January 10th t…
    Solitude and Society - by Vivian Loh - circles
  • I was honestly so stressed about CMM for the last couple of weeks/months before CMM. Since there were no higher-ups, if something went wrong then it really was just my fault. The entire competition was basically just tons of bullet points in my head and in piles of notes, AKA my DM’s with the Discord LaTeX bot. Unfortunately, until mid-December I had been completely locked in on fall-term academics (basically the way Caltech is, you spend ALL your time on problem sets and barely have time to sleep let alone other things…) and basically I had done on CMM was secure sponsorships and reserve most…
    Solitude and Society - by Vivian Loh - circles
  • And these ideas are perennial because they never seem to happen. It’s like the Solow paradox on an individual level. Why? I think there are two reasons: first, most people lack what Andy Matuschak calls a “serious context of use” (AI doesn’t move the needle because there’s no needle to move); second, most people are bottlenecked by internal factors where AI (or anything external, for that matter) can’t move the needle.
    Human Bottlenecks
  • See, most of us, unless we are students, we really don’t have cause to take notes on anything. If you’re a student, you take notes from the textbook. I keep a journal, which is occasionally useful. At some jobs I’ve kept a work journal, this has also been useful. If I stopped, probably, not much would change.
    Human Bottlenecks
  • And on top of the personal need, there has to be resistance. You have to be wanting the thing and not getting it. It is the feeling of “Why don’t they just ____?” or “Why can’t they ____?” in the face of a reality where they clearly aren’t doing or being those things. It is an awkward stalemate, a desire locked in suspension.1 But to cling to resentment is to cling to your own helplessness. If you maintain the belief that it is on others to change to make you happy, then the burden doesn’t fall on you. You can stay the same, and wait for what you want to fall out of the sky. You can stay in a …
    Resentment is needing others to change so you can stay the same
  • Even “the details of sex-announcing behaviors… contribute to the reduction of women and the elevation of men”[49]—men offering a firm handshake to other men, for example, but not to women. We live in a culture that has learned to recognize many of these habits as problematic, and that in least in some quarters is working hard to fashion new habits. But they have not gone away, and on Frye’s account, they are far more important than we realize to the reproduction of gender injustices further up the structural ladder.
    Microaggressions as Violence | PracticalMattersJournal
  • microaggressions should be understood as the patterns of personal interaction that perpetuate oppression
    Microaggressions as Violence | PracticalMattersJournal
  • Structural violence can take a wide variety of forms. It can exist at any level of organization: a family that denies daughters the opportunities offered to sons, a business that makes it impossible for an employee to report cases of harassment, a criminal justice system that disproportionately penalizes certain crimes, an intergovernmental body that imposes unequal economic burdens on different members. And it can be mediated in any number of different ways: written laws, economic incentives, city planning, cultural norms. There is no single template for how structural violence works. We can …
    Microaggressions as Violence | PracticalMattersJournal
  • And as Cynthia Moe-Lobeda notes, structural violence is especially hard to see for those who benefit from it.
    Microaggressions as Violence | PracticalMattersJournal