Excellent post on how TikTok became what it is.
The Sorting Hat is perhaps the most curious plot device from the Harry Potter universe. Is it a metaphor for genetic determinism? Did Draco have any hope of not being a Slytherin? By sorting Draco into that house, did it shape his destiny? Is the hat a metaphor for the U.S. college admissions system, with all its known biases? Is Harry Potter, sorted into Gryffindor, a legacy admit?
Instagram is some strange hybrid mix of social and interest graph, and now it’s also a jumble of formats, with a Stories feed relegated to a top bar in the app while the more stagnant and less active original feed continues to run vertically as the default. Messaging is pushed to a separate pane and also served by a separate app. Longer form videos bounce you to Instagram TV, which is just an app for videos that exceed some time limit, I guess? And soon, perhaps commerce will be jammed in somehow? Meanwhile, they have a Discover tab, or whatever it is called, which seems like it could be the default tab if they wanted to take a more interest-based approach like TikTok. But they seem to have punted on making any hard decisions for so long now that the app is just a Frankenstein of feeds and formats and functions spread across a somewhat confused constellation of apps.
See tweet section for more Chrisopher Alexander
By Tristan Hume
Fun series of lectures I’ve been watching when I have some free time and feel like thinking a bit but not too hard. Companion website
- adversarial examples often transfer between models - leverage three key ideas: 1. "we use the recent black box adversarial example construction techniques that create adversarial examples for a target model without access to the model’s architecture or parameters" 2. "we adapt machine learning models to mimic the initial visual processing of humans, making it more likely that adversarial examples will transfer from the model to a human observer" 3. "we evaluate classification decisions of human observers in a time-limited setting, so that even subtle effects on human perception are detectable" - "We find that adversarial examples that transfer across computer vision models do successfully influence the perception of human observers, thus uncovering a new class of illusions that are shared between computer vision models and the human brain" - Adversarial examples that transfer between models also affect human response time and accuracy
“We build and maintain an open repository of web crawl data that can be accessed and analyzed by anyone.”
Apparently neopanamax and post-neopanamax are ship size standards.
Abstract: The method consists in extracting of soluble substances in dried tea leaf, containing in a porcelain or earthenware pot, by means of freshly boiling water, pouring of the liquor into a white porcelain or earthenware bowl, examination of the organoleptic properties of the infused leaf, and of the liquor with or without milk or both.
A -> Bvs
observe[A -> B]