Franklyn Wang
23 followers · 5 following · 1686 views
on the atlas — 43
- Calvin and Hobbes at Martijn's - Bill Watterson30 savers
- Reality has a surprising amount of detail49 savers
- How factories were made safe2 savers
- What Gen Z Got for Christmas in 20231 savers
- Revised Psychology of Human Misjudment1 savers
- Understanding The Web Security Model, Part IV: Cross-Origin Resource Sharing (CORS)2 savers
- Log Structured Merge Trees1 savers
- [2309.07062] Large Language Models for Compiler Optimization1 savers
- A friendly introduction to machine learning compilers and optimizers10 savers
- In This Essay I Will: On Distraction3 savers
- Thoughts about what worked in math circles1 savers
- Yan Zhang on Twitter: "Thread re: @TheOtherHaider to untangle a very personal traumatic experience that I am still working to overcome. tl;dr: I have no regrets, except this single experience that represents a force in the world that is powerful, destructive, and unnerving. 1/" / Twitter3 savers
- Constitutional AI: RLHF On Steroids2 savers
- [2303.18231] Programmed but Arbitrary Control Minimization of Amplitude and phase for speckle Nulling (PACMAN)1 savers
- Beware Systemic Change | Slate Star Codex4 savers
- The Waluigi Effect (mega-post) - LessWrong16 savers
- The Invention of EBITDA2 savers
- A Mathematical Framework for Transformer Circuits39 savers
- Book Review: The Man From The Future - by Scott Alexander2 savers
- Still Alive1 savers
- Robert Nishihara on Twitter: "Exciting to see Quokka at the top of Hacker News (written by Ziheng Wang). In ~1000 lines of Python, Quokka is a high performance fault-tolerant query engine built on 1⃣ Ray (@raydistributed) - distributed execution 2⃣ Polars - fast dataframes 3⃣ Arrow (@ApacheArrow) - fast I/O https://t.co/khz7rLYqtN" / Twitter1 savers
- Neurotic Gradient Descent: (mis)Translating the Buddha5 savers
- Being Bored is Better than Social Media3 savers
- A year of new avenues18 savers
- Awaken The Giant Within1 savers
- American Idle — Remains of the Day5 savers
- Deep Listening at the Recurse Center6 savers
- advice - nabeelqu43 savers
- Reflections on a decade of coding5 savers
- Game Emulation via Neural Network2 savers
- Ruler: Inferring Rewrite rules with Equality Saturation1 savers
- California can't save every burned town1 savers
- The Internet Was Designed With a Narrow Waist1 savers
- Good conversations have lots of doorknobs18 savers
- People Spend Too Much Time On Decisions with Equally Satisfying Outcomes11 savers
- Potential fabrication in research threatens beta-amyloid theory5 savers
- Status as a Service (StaaS) — Remains of the Day27 savers
- A Big Little Idea Called Legibility15 savers
- The Lesson of Grace in Teaching, by Francis Su12 savers
- Just Ask for Generalization | Eric Jang10 savers
- Up and Down the Ladder of Abstraction10 savers
- The Most Important Scarce Resource is Legitimacy8 savers
- amirgamil/apollo: A Unix-style personal search engine and web crawler for your digital footprint.2 savers
highlights — 45
In other words, simple requests are (approximately) those you could do without CORS, which means that they are safe to do with CORS, as long as the server agrees to the JS having access to the data.
Understanding The Web Security Model, Part IV: Cross-Origin Resource Sharing (CORS)as in most cases the attacker can just connect directly to the server and retrieve the data
Understanding The Web Security Model, Part IV: Cross-Origin Resource Sharing (CORS)The fugitive moments in between our lifelong undertakings, whatever their ultimate worth, may be what we are searching for all along. Maybe we are distracted because we are still learning how to live.
In This Essay I Will: On DistractionGeometry was a complete dud, because it's even harder to come up with stories about arbitrary figures with shaded regions you want to find the area of.
Thoughts about what worked in math circlesFor instance, what if we got rid of medical malpractice lawsuits in favor of a no-fault system in which the medical provider always pays for any complication arising under their care?
How factories were made safeMy conclusion is that safety goals will rarely if ever be driven bottom-up. They are much more likely to be driven top-down, by someone in a position of leadership or authority who sees the statistical picture and can implement the systematic change, including education and training, needed to make it happen.
How factories were made safeApplying the mixed product property and collapsing identities yields:
A Mathematical Framework for Transformer CircuitsBerlin street scene. World War I: man at corner yelling repeatedly: “The Kaiser is an idiot!” Out of nowhere two police agents appear and arrest him for high treason. “But I was referring to the Austrian Kaiser, not to our Kaiser.” “You can't fool us! We know who the idiot is.”
Book Review: The Man From The Future - by Scott AlexanderI tried to correct him, but it seemed like his email client only accepted replies from people on his contact list.
Still Alive"Quokka is not fault tolerant, though it will be by the end of 2022. This is how I intend to be collecting my PhD, so you can be pretty darn sure it will happen."
Robert Nishihara on Twitter: "Exciting to see Quokka at the top of Hacker News (written by Ziheng Wang). In ~1000 lines of Python, Quokka is a high performance fault-tolerant query engine built on 1⃣ Ray (@raydistributed) - distributed execution 2⃣ Polars - fast dataframes 3⃣ Arrow (@ApacheArrow) - fast I/O https://t.co/khz7rLYqtN" / TwitterIn Buddhist psychology, the victim-oppressor mindset is called the Hell Realm because it is considered a particularly nasty maladaptive strategy. Not only because it is miserable for people caught in it, but because it reinterprets signs explaining how to get out as tricks, attacks, etc. (See: my frustration with wireheading objections ;) It is said that the most tragic aspect of the hell realm is that none of the gates are barred.
Neurotic Gradient Descent: (mis)Translating the BuddhaBut as I've mentioned elsewhere, we do know from animal behavioral studies that positive reinforcement just works way better than punishment. Punishment tends to cause generalized avoidant mechanisms to kick in, which is why people experience akrasia.
Neurotic Gradient Descent: (mis)Translating the BuddhaBecause others still live within the paradigm of seeing everything as related to houses, they are inclined to perceive you as living in a shiny spiritual house, and try to figure out which sets of maintenance activities and decorations will grant them these ‘spiritual’ advantages.
Neurotic Gradient Descent: (mis)Translating the BuddhaWe ignore or try to push away aspects of the objects that make us feel hopeless, helpless, worthless, empty. This carving up of objects has little to do with how they are in the world, so the mental stories we build out of these carved up objects are also incoherent.
Neurotic Gradient Descent: (mis)Translating the BuddhaYou currently experience obstacles in the course of pursuing some goal as stressful. In order to generate the necessary energy to overcome the stressor you generate a mental construct that causes you to suffer even more if you don’t overcome the stressor. So when people imagine a decrease in mind created stress, they imagine only the secondary motivation-hack stress going away after which they will become useless in the face of any mild obstacles in life (just go with the flow, man!). Instead what happens is that both kinds of stress decrease at the same time.
Neurotic Gradient Descent: (mis)Translating the BuddhaWhen one experiences a lessening of Tanha, the objection “but what if by stressing out just a bit more some great non-linear results would have been realized in your life that were totally worth it!” starts to sound like “but what if being tense at all times just happened to be exactly what kept you from getting hit by that random bus?”
Neurotic Gradient Descent: (mis)Translating the BuddhaA kind of happiness that comes from a halting of believing that we need to get happiness by arranging things to match up with mental projections.
Neurotic Gradient Descent: (mis)Translating the BuddhaIn short, that Nicca, Atta, and Sukha (the opposite of Dukkha) are maladaptive strategies. Not only do they not get us what we want but they maintain the conditions that lead us to keep using them.
Neurotic Gradient Descent: (mis)Translating the BuddhaWhat Anatta should be taken to mean in relation to philosophy of science is more like ‘you should never have a prior that is non-updateable by sense data.’
Neurotic Gradient Descent: (mis)Translating the BuddhaAgain, the point is to spot the mental moment of playing with the fake steering wheel instead because it is easier.
Neurotic Gradient Descent: (mis)Translating the BuddhaAnd this gets at an extremely important point in how this stuff works. If suffering were truly just coming in from the outside in thousands of different forms (i.e. the way things seem on cursory inspection) then we wouldn't have much hope of a single intervention helping us. Nor would we be confident in any such intervention since some new form of suffering can always show up.
Neurotic Gradient Descent: (mis)Translating the BuddhaSo Nicca is our tendency to believe that things could or should be maintained to our satisfaction. This is an identifiable mental event in how we reify an object or concept.
Neurotic Gradient Descent: (mis)Translating the BuddhaHow will the world now be reshaped to fit their needs?
A year of new avenues(Today, certain container ports are among the most automated spaces on earth, with bright lines painted on the ground, indicating zones of machine activity. If you step across, there’s a good chance you’ll get squished.)
A year of new avenuesLimited references create a limited life. If you want to expand your life, you must expand your references by pursuing ideas and experiences that wouldn’t be a part of your life if you didn’t consciously seek them out.
Awaken The Giant WithinPeople with an impoverished vocabulary live an impoverished emotional life; people with rich vocabularies have a multi-hued palette of colors with which to paint their experience, not only for others, but for themselves as well.
Awaken The Giant WithinAll goal setting must be immediately followed by both the development of a plan, and massive and consistent action toward its fulfillment.
Awaken The Giant Withinanything you want that’s valuable requires that you break through some short-term pain in order to gain long-term pleasure.
Awaken The Giant WithinAdditionally, it seems like the more secure one feels in their life, the better one can listen to someone else, probably because the troubles of life aren't unconsciously eroding one's attention. When I'm listening well, I feel a meditative calmness that seems to reach down into my nervous system.
Deep Listening at the Recurse Centermost people fail more from many days of zero output than they do from not maximizing output on any given day
advice - nabeelquIn my case, this just meant I needed to occasionally have the network inputs be previous neural network outputs (instead of “real” frames) during training, so that the network would learn to continue gameplay from a previous network-predicted frame.
Game Emulation via Neural NetworkTo encourage it to produce correct reasoning steps, the Inference model is trained to predict an en- tailment given only the selection. By not allowing the Inference model access to the question, we prevent it from “cheating” (directly predicting the answer from the question). While we cannot guar- antee that every reasoning step is correct, in the sense that the inference logically follows from the selection (Defn. 4), our implementation makes this more likely. Under the assumption that the Inference model produces logically correct infer- ences, our model is guaranteed to produce valid reaso…
Faithful Reasoning (with LLMs)What matters most, then, is not how much we give or take, but whether we offer and accept affordances. Takers can present big, graspable doorknobs (“I get kinda creeped out when couples treat their dogs like babies”) or not (“Let me tell you about the plot of the movie Must Love Dogs…”).
Good conversations have lots of doorknobsThe brain goes nuts trying to find order in the chaos. Instead, the researcher says, they usually start with something like a black-and-white checkerboard pattern.
A Big Little Idea Called LegibilityThe picture is not an exception, and the word “legibility” is not a metaphor; the actual visual/textual sense of the word (as in “readability”) is what is meant. The book is full of thought-provoking pictures like this: farmland neatly divided up into squares versus farmland that is confusing to the eye, but conforms to the constraints of local topography, soil quality, and hydrological patterns; rational and unlivable grid-cities like Brasilia, versus chaotic and alive cities like Sao Paolo.
A Big Little Idea Called LegibilityWe make this mistake because we are tempted by a desire for legibility.
A Big Little Idea Called Legibility…most people will agree that the statements accurately describe them. In fact, most people will feel like they’re unusually accurate descriptions, which is how astrologers get you.
Forer Statements As Updates And AffirmationsGrace gives people dignity they don’t have to earn.
The Lesson of Grace in Teaching, by Francis SuThis idea is satirized in the philosophical paradox of Buridan's ass, where a donkey that is equally hungry and thirsty is placed exactly halfway between a stack of hay and a bucket of water. Unable to choose between them, it dies.
People Spend Too Much Time On Decisions with Equally Satisfying OutcomesUnable to choose between them, it dies.
People Spend Too Much Time On Decisions with Equally Satisfying OutcomesThe anomalies with Newtonian gravity weren’t things like “action at a distance doesn’t feel scientific enough” or “it doesn’t sufficiently glorify Jesus Christ” or even “it’s insufficiently elegant”. The one that ended up most important was “its estimate for the precession of the orbit of Mercury is off by forty arc-seconds per century”.
Criticism Of Criticism Of Criticism - by Scott AlexanderThe question we ask is: given an explicit limit K on the number of video frames we can jointly model, how can we best allocate these frames to generate a video of length N > K ? One option is to use the previously-described autoregressive model but, if K = N/ 4 , we could instead follow Ho et al. [15] by training two models: one which first samples every 4th frame in the video, and another which (in multiple stages) infills the remaining frames conditioned on those.
Flexible Diffusion Modeling of Long VideosThe idea is that given a heavy-tail distribution for flow sizes, the more significan t bits in the counters are poorly utilized
counter.dvi“The current paradigm of pain management focuses mostly on treatment of a physical origin of pain, however, in many cases of chronic back pain a physical source of pain cannot be identified,” said corresponding author Michael W. Donnino, MD, a physician in the Departments of Critical Care and Emergency Medicine at BIDMC. “Our group focused on the hypothesis that non-specific back pain is the symptomatic manifestation of a psychological process, substantively driven by stress, repressed emotions and other psychological processes. The exact mechanism remains unclear, but an analogy could be made…
Researchers’ novel mind-body program outperforms other forms of treatment for chronic back pain | BIDMC of BostonSome researchers out there are probably pursuing this already. My guess is that the research community tends to reward narratives that increase intellectual complexity and argue that “we need better algorithms”. People pay lip service to “simple ideas” but few are willing to truly pursue simplicity to its limit and simply scale up existing ideas.
Just Ask for Generalization | Eric Jang