flâneur

Markus Strasser

4 followers · 13 following · 576 views

on the atlas — 26

highlights — 49

  • Whoever shall put away his wife, let him give her a writing of divorce
    The Gospel of Matthew
  • You know, secretly, even if you’re pretending not to, that this thing is nearing exhaustion. There is simply nothing there online. All language has become rote, a halfarsed performance: even the outraged mobs are screaming on autopilot. Even genuine crises can’t interrupt the tedium of it all, the bad jokes and predictable thinkpieces, spat-out enzymes to digest the world.
    The internet is already over - by Sam Kriss
  • ‘Boredom is the dream bird that broods the egg of experience.’ But when I’m listlessly killing time on the internet, there is nothing. The mind does not wander. I am not there. That rectangular hole spews out war crimes and cutesy comedies and affirmations and porn, all of it mixed together into one general-purpose informational goo, and I remain in its trance, the lifeless scroll, twitching against the screen until the sky goes dark and I’m one day closer to the end. You lose hours to—what? An endless slideshow of barely interesting images and actively unpleasant text. Oh, cool—more memes! Yo…
    The internet is already over - by Sam Kriss
  • ‘Do you know why the Chinese are so naturally good at deep learning? Because the black box has been part of Chinese society and Chinese culture since the very beginning
    Architecting Reality
  • A superpower encodes the game, a normal power plays the game.
    Architecting Reality
  • This is the life we’ve chosen, and we like it.
    A glimpse of the other side - by mingyuan - bright distance
  • It can be compared, diffed, clustered, corrected, summarized and filtered algorithmically. It permits multiparty editing. It permits branching conversations, lurking, annotation, quoting, reviewing, summarizing, structured responses, exegesis, even fan fic. The breadth, scale and depth of ways people use text is unmatched by anything. There is no equivalent in any other communication technology for the social, communicative, cognitive and reflective complexity of a library full of books or an internet full of postings. Nothing else comes close. So this is my stance on text: always pick text fi…
    Always bet on text
  • it's vastly cheaper on a symbol-by-symbol basis
    Always bet on text
  • Text can convey ideas with a precisely controlled level of ambiguity and precision, implied context and elaborated content, unmatched by anything else. It is not a coincidence that all of literature and poetry, history and philosophy, mathematics, logic, programming and engineering rely on textual encodings for their ideas.
    Always bet on text
  • Text is the most flexible communication technology. Pictures may be worth a thousand words, when there's a picture to match what you're trying to say.
    Always bet on text
  • The implication of this simple example is that most of the “features” Lisp has are implemented in Lisp itself and can be reduced down to a small set of core features. Structures, object-oriented programming or loops can be simply added to Lisp. Compare this to other languages: if there is feature you want you first have to convince the standards committee that it is a good idea, then wait for the committee to agree upon an implementation and publish a new standard (and most likely you will have to wait for the committee to schedule a meeting in the first place), then you have to wait for imple…
    Programmable Programming Language
  • The best we can do is modeling a generic worm - pretraining and running the neural network with fixed weights. Thus, no worm is "uploaded" because we can't read the weights, and these simulations are far from realistic because they are not capable of learning. Hence, it's merely a boring artificial neural network, not a brain emulation.
    Whole Brain Emulation: No Progress on C. elgans After 10 Years - LessWrong
  • This seems like a research area where you have multiple groups working at different universities, trying for a while, and then moving on. None of the simulation projects have gotten very far: their emulations are not complete and have some pieces filled in by guesswork, genetic algorithms, or other artificial sources.
    Whole Brain Emulation: No Progress on C. elgans After 10 Years - LessWrong
  • Don’t just do trendy stuff. If something is really popular, I tend to think: back off. I tell myself and my students to go with your own aesthetics, what you think is important.
    Donald Knuth on work habits, problem solving, and happiness
  • A person’s success in life is determined by having a high minimum, not a high maximum. If you can do something really well but there are other things at which you’re failing, the latter will hold you back.
    Donald Knuth on work habits, problem solving, and happiness
  • In a sense I tend now to suspect that it was necessary to leave the Garden of Eden. Imagine a world where people are in a state of euphoria all the time — being high on heroin, say. They'd have no incentive to do anything.
    Donald Knuth on work habits, problem solving, and happiness
  • I save considerable time by reading several dozen papers on the same topic all in the same week, rather than reading them one by one as they come out and trying to keep infinitely many things in my head all at once.
    Donald Knuth on work habits, problem solving, and happiness
  • On the importance of anthropomorphizing a problem. "Another aspect of role playing is considerably more important: We can often make advances by anthropomorphizing a problem, by saying that certain of its aspects are "bad guys" and others are "good guys," or that parts of a system are "talking to each other." This approach is helpful because our language has lots of words for human relationships, so we can bring more machinery to bear on what we're thinking about."
    Donald Knuth on work habits, problem solving, and happiness
  • They should use their expertise now, while they have this unique ability, because they're going to lose it in a month. I emphasize that they shouldn't be satisfied with solving only one problem;
    Donald Knuth on work habits, problem solving, and happiness
  • Then finally they get to a harder problem, where the only way to solve it is with algebra. But by that time, they haven’t learned the fundamental ideas of algebra. The fact that they were so smart prevented them from learning this important crutch that I think turned out to be important for the way I approach a problem. Then they say, “Oh, I can’t do math.” They do very well as biologists, doctors and lawyers.
    Donald Knuth on work habits, problem solving, and happiness
  • And I'll use up 20 sheets or more per hour when I'm exploring a problem, especially at the beginning. For the first hour I'm trying all kinds of stuff and looking for patterns. Later, after internalizing those calculations or drawings or whatever they are, I don't have to write quite so much down, and I'm getting closer to a solution. The best test of when I'm about ready to solve a problem is whether or not I can think about it sensibly while swimming, without any paper or notes to help out. Because my mind is getting accustomed to the territory, and finally I can see what might possibly lead…
    Donald Knuth on work habits, problem solving, and happiness
  • I don't know if I ever would have learned to program if I had modern internet.
    Productivity advice - by Slava Akhmechet - Zero Credibility
  • I have a work computer, and a router with parental controls that blocks every possible internet distraction on it. No Twitter, no Hacker News, no YouTube. Router administration is set up so I can't make changes over WiFi. If I want to unblock something, I have to physically get to the router and plug in a cable to change the settings.
    Productivity advice - by Slava Akhmechet - Zero Credibility
  • Work means sitting down, getting through that calculus chapter, and doing the exercises. No amount of productivity hacking will make that easier. You don't need pomodoro alarms, bullet journals, time tracking apps, animated explainer videos, or different color highlighters. Everyone doesn't learn differently. Everyone learns calculus in the same way— by doing the work. You need Rudin's book, a pen, paper, and time. More tools give you negative utility. They won't make the work go faster. But they will consume as much time as you are willing to waste.
    Productivity advice - by Slava Akhmechet - Zero Credibility
  • If the only thing you must do is the work, why is there so much productivity advice? Blog posts, courses, seminars, software? Because when there is demand, there is supply. Work is hard. People will latch onto anything to avoid doing it. The market is happy to oblige.
    Productivity advice - by Slava Akhmechet - Zero Credibility
  • Improving this 30% at the tool level is where most of the improvement can come from, not from trying to generalize 30% using incrementally better models. AlphaFold, the most impressive advancement in AI for biology to this day, can come to the resolution that is only as good as (or worse than) crystallography data. You technically can not get to 0.5Å atomic resolution with your model if your tool is 1Å (Å - angstrom). So if membrane proteins[**] crystalize poorly (and they do crystallize poorly), your model will underperform on all of the membrane proteins until you improve the tooling that me…
    So where are we with deep learning for biochem? — lada nuzhna
  • It will surely undermine your efforts if you are looking for mechanistic insights into biology - and this is a really big (and if you asked me - most important) class of problems in biology.
    So where are we with deep learning for biochem? — lada nuzhna
  • Molecular tools still bottleneck the resolution of things My personal biological red pill happened when I learned that scRNA-seq captures only 30% of transcripts in the cell in the best-case scenario. The most informative readout of a cell’s state captures only 30% of its state! One might think it is not that bad if cells come from a homogenous population (which won’t be the case if you work with cancer by the way) and you have enough of them – the measurements are more robust against noise as you increase the number of cells. But it still means that the signal from every individual cell is we…
    So where are we with deep learning for biochem? — lada nuzhna
  • There are domains of biology where it isn’t all that bad: for example, the amino-acid sequence and its Coulombic map can be used to more or less reconstruct the structure of the protein. You can isolate the folding problem from the cell context around protein and succeed… unless you are trying to predict the folding of a protein given some co-enzymes or ligands, or as part of a protein complex, or under different pH… or some other realistic scenario. In this case, we come back to the same conclusion – if you can not make some crude assumptions, most problems in biology aren’t easily reducible …
    So where are we with deep learning for biochem? — lada nuzhna
  • This is also the reason you can't really translate the success of image processing into biology or chemistry. When you get an image, more often than not it contains all the information you need to derive a specific conclusion. Working with molecules, even when powered with deep learning and machine learning, is non-trivial because half of the information is already missing from the dataset before you even started your work. This sets an upper limit of accuracy of the type of patterns we can extract.
    So where are we with deep learning for biochem? — lada nuzhna
  • Although it has been a hit in pharmacology for a while, today, almost 2 decades later, we see that Lipinski rules fail to comply with almost half of the approved drugs, with researchers calling into question whether a thing called “drug-like properties” even exists. Criticism, in this case, is similar – you can not ignore the molecular context (e.g., availability of transporters to deliver your drug) to predict drug properties. And if you hope to generalize heuristics from the already approved drugs, you will likely miss many important leads (just like it happened with the rule of five).
    So where are we with deep learning for biochem? — lada nuzhna
  • I spent the next couple years unemployed, working with my physicians to try and recover my health while occasionally writing code. I’m happy to report that I’m partially recovered at this point and being paid to work on open source, but I’ll never be the same.
    Why I Quit Google’s WebAssembly Team, And How It Made Me Sick | by Katelyn Gadd | May, 2022 | Medium
  • Google is the worst place I’ve ever worked and it quite literally gave me brain damage.
    Why I Quit Google’s WebAssembly Team, And How It Made Me Sick | by Katelyn Gadd | May, 2022 | Medium
  • In her essay Regarding the Pain of Others, on images of violence and suffering, Susan Sontag famously asks: “What is the point of exhibiting these pictures? To awaken indignation? To make us feel ‘bad,’ that is, to appall and sadden? To help us mourn? … Do they actually teach us anything?” This question — what is the point — has been on my mind a lot lately. As I scroll through TikTok and find images of war, tear-stained confessionals, and cooking videos airing generational wounds over aesthetic s’mores; read novels where characters are put through gauntlets of torture befitting Job; watch sho…
    Hard to See — Real Life
  • We argue that the “node and edge-centric” mindset of current graph deep learning schemes imposes strong limitations that hinder future progress in the field. As an alternative, we propose physics-inspired “continuous” learning models that open up a new trove of tools from the fields of differential geometry, algebraic topology, and differential equations so far largely unexplored in graph ML.
    Beyond Message Passing: a Physics-Inspired Paradigm for Graph Neural Networks
  • Learning as an optimal control problem The space of all possible states of a process at a given time can be regarded as a hypothesis class of functions to be represented. Framed this way, learning is posed as an optimal control problem [46]. The goal becomes to control the process, by choosing a trajectory in the parameter space, to reach a certain desirable state. Using optimal control terminology, expressive power can be formulated as how well a given function can be reached by a trajectory in the parameter space. Efficiency is related to how long it takes to reach a certain state. Finally, …
    Beyond Message Passing: a Physics-Inspired Paradigm for Graph Neural Networks
  • This is also why we will see more and more companies choosing employees based on increasingly specific skills and traits, not just professional traits, but also personal ones such as values, political views, gender, ethnicity, and more.
    Rise of the 10X Class
  • The internet makes it possible to monetize increasingly narrow niches, and it drives customers to expect solutions that are increasingly tailored to their needs and aspirations. These solutions are developed and delivered by people with the specific skills and traits that matter to that niche.
    Rise of the 10X Class
  • As you recall, when “imperfect substitution” is combined with “joint consumption technologies,” the result is “the possibility for talented persons to command both very large markets and very large incomes.“ While music and TV stars can be broadcast from and to anywhere, software engineers can only work in one place.
    Rise of the 10X Class
  • The number of potential employers and, thus, the total income of the most productive employees is capped due to geography constraints.
    Rise of the 10X Class
  • The internet makes it possible for many knowledge employees to work from anywhere. The earning potential of (many of) the most productive employees is no longer capped by geography. As a result, we will see the emergence of a new class of people earning salaries that are an order of magnitude higher than what we saw in previous decades.
    Rise of the 10X Class
  • One top-notch engineer is worth "300 times or more than the average," explains Alan Eustace, a Google vice president of engineering. He says he would rather lose an entire incoming class of engineering graduates than one exceptional technologist. Many Google services, such as Gmail and Google News, were started by a single person, he says.
    Rise of the 10X Class
  • To get back on track, we must quit our comfortably lazy routines and leap back into the unknown wilderness. We must first set our curiosity and then our fatal determination on the biggest problems of our collective existence and functional justice, without assurance that we will get it right.
    Quit Your Job
  • Where will you get ideas? You’ll find them out on the un-tracked frontier while playfully chasing hunches for novel value. You get good ideas from years of hard leisure.
    Quit Your Job
  • There are investments you can’t make from a structured, nine-to-five, narrowly teleological environment. You have to let your life go fallow sometimes, like a crop rotation giving the land time to bring forth new fertility. This is actually a consequence of a fairly general theorem about how to find treasure in complex search spaces: The best search strategies for complex problems like life generally don’t seek out particular homogeneous objectives, but interesting novelty. The search space is too complicated and unknown for linear objective-chasing to work. Biological evolution, in practice, …
    Quit Your Job
  • Contemporary American governance is plagued by a surplus of grifters, offering a disparate array of optimizations, marginal improvements, and success metrics. This is policy done by wonks, not visionaries. San Francisco must have a clear and bold vision of what it must become. Policy should be the result of reverse-engineering that outcome to figure out the path to success. A prosperous and growing metropolis demands density and its benefits, which in turn requires that the fixed land supply be properly managed.
    San Francisco's Future Should Begin with a Land Value Tax
  • A society of masters is far stronger than a society of narrowly focused experts. The ancient Athenian state privileged this kind of social order; numerous important positions were filled by lottery from among pools of Athenian citizens and it was expected that they could perform such social duties when chosen. Is there any major institution today that has this level of faith in its members?
    The Rebirth of Industrial Mastery
  • Don't try to go too fast for those things. Learning mathematics and physics properly is extremely hard. I'm not saying you need to do it for every single life path, but I'm saying that if you want to do something very hard in the physical world, there are prerequisites you can’t ditch—you have to go through them.
    Master Plan - Justin Glibert (Foundation) - by David
  • I really don't think people can go and take five courses at a university and read 5 million blog posts and then build a company around that.
    Master Plan - Justin Glibert (Foundation) - by David