flâneur

Tazik Sh

17 followers · 29 following · 1268 views

on the atlas — 179

highlights — 316

  • I am a big fan of pseudoscience. One reason is I believe most novel scientific concepts were first dismissed as being ridiculous "psuedoscience". Imagine being the guy who first thought little invisible bacteria were making us sick while everyone else was
    Magical Beans | Andy Kong
  • Nonviolence. One of the lessons of the last 200 years of social change is that nonviolence— a commitment to compassion even for one’s adversaries—beats violence
    We Pulled Off an “Impossible” Rescue. Here’s How We’ll Do It Again.
  • The influence of one token upon another is determined by the QKTQK^TQKT dot product - so that's exactly where we should focus our positional encoding!
    You could have designed state of the art positional encoding
  • So the encoding scheme designed by Noam Shazeer in Attention is all you need was already encoding relative position as a rotation back in 2017
    You could have designed state of the art positional encoding
  • Anki is a meaning-making context. It is a near universal context for caring.
    How I use memory systems
  • The system made the mayor of every major city personally and politically responsible for the production, supply, and price stability of non-staple foods for their citizens.
    China's "Vegetable Basket Program": How a Nation Achieved Food Abundance - PandaYoo
  • This symbiotic relationship—where a central government mandate creates a high-pressure environment that incentivizes local officials to find, nurture, and scale up grassroots innovations—is a recurring theme in China’s economic development.
    China's "Vegetable Basket Program": How a Nation Achieved Food Abundance - PandaYoo
  • demonstrating the potential of wearable ultrasound in digestive health monitoring.
    Wearable ultrasound technology | Nature Reviews Bioengineering
  • After my mother died, I wished I'd spent more time with her. I lived as if she'd always be there. And in her typical quiet way she encouraged that illusion. But an illusion it was. I think a lot of people make the same mistake I did.
    Life is Short
  • When writing, your first job is this: First, make me care.
    First, Make Me Care, by Gwern · Gwern.net
  • intuition is fake! Really good resources make you feel like you understand something, but to actually understand something, you need to engage with it
    Building intuition with spaced repetition systems | Jacob’s Blog
  • SF people are notoriously flakey and slimy and don’t care if you aren’t building something.
    do I want to "make it" or do I just want to be held
  • By 1971, her team had screened over 2,000 traditional Chinese recipes and made 380 herbal extracts, from some 200 herbs, which were tested on mice.[10] One compound was effective, sweet wormwood (Artemisia annua)
    Tu Youyou
  • I don't understand how the experimental setup provides evidence for self-other overlap working.
    Self-Other Overlap: A Neglected Approach to AI Alignment — LessWrong
  • use your allies on the bottom (front-line workers) and the top (executives) to squeeze out your opponents in the middle (managers, often in IT or data science departments, who would rather not have an external vendor encroach on their turf).
    The Great Data Integration Schlep - by Sarah Constantin
  • only sell to companies facing an “existential threat”, i.e. where something has gone so drastically wrong they might go out of business altogether
    The Great Data Integration Schlep - by Sarah Constantin
  • What is remarkable is that they have invested in a large number of people — at least a third of the company by headcount — to descend en masse to the customer’s facilities, set up and customize their database and associated tools, teach the users to work the UI, and, crucially, negotiate for data access.
    The Great Data Integration Schlep - by Sarah Constantin
  • When you’re a software vendor to a large company, it’s not at all unusual for it to be easier to make a multi-million dollar sale than to get the data access necessary to actually deliver the finished software tool.
    The Great Data Integration Schlep - by Sarah Constantin
  • In a manufacturing context, a lot of important data is not on computers.
    The Great Data Integration Schlep - by Sarah Constantin
  • Instead, we hypothesize that there are a large number of configurations (in weight space) of foundational models with the same or similar capabilities and that by introducing self-other overlap training we can push the learning process away from models with deceptive capabilities.
    Self-Other Overlap: A Neglected Approach to AI Alignment — LessWrong
  • All men dream: but not equally. Those who dream by night in the dusty recesses of their minds wake up in the day to find it was vanity, but the dreamers of the day are dangerous men, for they may act their dreams with open eyes, to make it possible.
    Telescoping Ambition
  • No book is ever going to tell me what I need to do, and I can’t substitute another man’s ambition for my own.
    Telescoping Ambition
  • So I think the way to deal with your parents' opinions about what you should do is to treat them like feature requests. Even if your only goal is to please them, the way to do that is not simply to give them what they ask for. Instead think about why they're asking for something, and see if there's a better way to give them what they need.
    Module 1: Deciding to start a startup | Startup School - The Best Resource for Founders
  • They enslave their children’s children who make compromise with sin.
    Unsong
  • Everybody knows making your bed unlocks a bunch of other positive feedback loops. My friend Jessica says in her experience, "solving" doing the dishes does the same for group houses. These things where if you do them, a cascade of other things fall into place.
    On Grace? - by xiq - Fractal Sidequests
  • A language model aims to learn, from the sample text, a distribution close to the empirical distribution of the language.
    Evaluation Metrics for Language Modeling
  • The way to test an idea is to either launch it and see what happens or try to sell it (e.g. try to get a letter of intent before you write a line of code.) The former works better for consumer ideas (users may tell you they will use it, but in practice it won’t cut through the clutter) and the latter works better for enterprise ideas (if a company tells you they will buy something, then go build it.)
    Startup Playbook
  • If you do not work on an important problem, it's unlikely you'll do important work. It's perfectly obvious.
    You and Your Research
  • drive, misapplied, doesn't get you anywhere. I've often wondered why so many of my good friends at Bell Labs who worked as hard or harder than I did, didn't have so much to show for it. The misapplication of effort is a very serious matter. Just hard work is not enough - it must be applied sensibly.
    You and Your Research
  • Once you get your courage up and believe that you can do important problems, then you can. If you think you can't, almost surely you are not going to.
    You and Your Research
  • The particular thing you do is luck, but that you do something is not.
    You and Your Research
  • the difference between those who do and those who might have done.
    You and Your Research
  • Years later, Potok still remembers a fellow attendee telling him about his plans to bioengineer shrimp to feel less pain.
    The Business (and Social Good) of Destroying Old Air Conditioners
  • Most people, when asked, say they want to do the most good for the world. But aiming for social good is not how you change the world. Aiming for curiosity is.
    Don't try to cure cancer — Ada Nguyen
  • Greatness is a matter of survival, and curiosity helps you stay in the fight.
    Don't try to cure cancer — Ada Nguyen
  • In many cases it's more accurate to model the world as 500 people than 8 billion
    Nat Friedman
  • A week is 2% of the year
    Nat Friedman
  • So here is the ultimate advice for young would-be startup founders, boiled down to two words: just learn.
    Before the Startup
  • Ok, so how do you turn your mind into the type that startup ideas form in unconsciously? (1) Learn a lot about things that matter, then (2) work on problems that interest you (3) with people you like and respect.
    Before the Startup
  • I wrote "become good at some technology." But that prescription, though sufficient, is too narrow. What was special about Brian Chesky and Joe Gebbia was not that they were experts in technology. They were good at design, and perhaps even more importantly, they were good at organizing groups and making projects happen. So you don't have to work on technology per se, so long as you work on problems demanding enough to stretch you.
    Before the Startup
  • You can do things in your early 20s that you can't do as well before or after, like plunge deeply into projects on a whim and travel super cheaply with no sense of a deadline. For unambitious people, this sort of thing is the dreaded "failure to launch," but for the ambitious ones it can be an incomparably valuable sort of exploration. If you start a startup at 20 and you're sufficiently successful, you'll never get to do it.
    Before the Startup
  • If you start a startup, it will take over your life to a degree you cannot imagine. And if your startup succeeds, it will take over your life for a long time: for several years at the very least, maybe for a decade, maybe for the rest of your working life. So there is a real opportunity cost here.
    Before the Startup
  • If you're thinking about getting involved with someone — as a cofounder, an employee, an investor, or an acquirer — and you have misgivings about them, trust your gut. If someone seems slippery, or bogus, or a jerk, don't ignore it.
    Before the Startup
  • That’s right, we can just Taylor expand the network function with respect to the weights around its initialization.
    Understanding the Neural Tangent Kernel – Rajat's Blog – A blog about machine learning and math.
  • Note that 𝑊[𝑥,𝑦]=𝑊1𝑥+𝑊2𝑦 𝑊 [ 𝑥 , 𝑦 ] = 𝑊 1 𝑥 + 𝑊 2 𝑦 where [ ] [ ] denotes concat and 𝑊 𝑊 is split horizontally into 𝑊1 𝑊 1 and 𝑊2 𝑊 2 . Compare this to 𝑊(𝑥+𝑦)=𝑊𝑥+𝑊𝑦 𝑊 ( 𝑥 + 𝑦 ) = 𝑊 𝑥 + 𝑊 𝑦 . So you can interpret adding as a form of concatenation where the two halves of the weight matrix are constrained to 𝑊1=𝑊2 𝑊 1 = 𝑊 2 .
    keras - When to "add" layers and when to "concatenate" in neural networks? - Cross Validated
  • Maybe one day, we'll be able to build environments where young people feel more free to pursue their ambitions, without the strong tides of society pulling them toward the path of certainty.
    The Breaking Point of Ambition
  • What’s an apologist, then? Someone who trusts their intuitions deeply and tries to investigate the underlying structures that give rise to those intuitions, rather than inventing some theory from first principles and overriding any instincts that disagree.
    Dreams of the Collective Unconscious - by BLAP
  • . This is because, I claim, everyone of sufficient age already knows everything important to be known about the human psyche.
    Dreams of the Collective Unconscious - by BLAP
  • Studies enter the popular awareness to the extent that their results are memetically powerful, and it’s the fact of being memetically powerful, not of being rigorous, that constitutes evidence about what human minds are and aren’t like.
    Dreams of the Collective Unconscious - by BLAP
  • Likewise, given how little flooding threatened the average person throughout history compared to other fears like disease and predators, it fascinates me how universal the flood myth is as a collective dream
    Dreams of the Collective Unconscious - by BLAP