currymj

38 ms·

451.
by currymj 6y ago
most US earthquakes are in California where Spanish is widely spoken and lots of things get Spanish names. however I don’t think this is common usage. it might be “headline-ese” as the word has fewer characters than earthquake.
452.
by currymj 6y ago
it's definitely possible that individuals using these drugs carefully and deliberately could discover some kind of valuable practice or insight. And perhaps this could turn into something that eventually could be shared among many peop
453.
by currymj 6y ago
i think if they really work well as part of treatment for PTSD and other aftereffects of trauma, and perhaps other mental illnesses, that would be of profound value for humanity. as far as people taking them in general: from looking at peop
454.
by currymj 6y ago
Stamets himself is quite a successful entrepreneur, as one data point. He funds much of his research out of the proceeds from his business selling mushroom-based products.
455.
by currymj 6y ago
i don't like this argument at all. protecting the environment is an immense responsibility that we all share, so the argument veers close to implying that choosing not to take psychedelic drugs is morally wrong -- willfully choosing to
456.
by currymj 6y ago
thanks for the good faith answer, this is reassuring and interesting. there’s a long history, starting way back with Timothy Leary, of psychedelic drug promoters doing sloppy, dubious science. but there’s also a growing orthodoxy in some ci
457.
by currymj 6y ago
is there any researcher who is really excited about the profound world-historical importance of fungi, but who has not taken psychedelic mushrooms? it would help me to be less skeptical of the rhetoric.
458.
by currymj 6y ago
economics and mechanism design is in the DNA of both Facebook and Google — core business of both is running auctions. that background tends to go with thinking prediction markets are cool.
459.
by currymj 6y ago
natural gradient descent is big in reinforcement learning -- proximal policy optimization (popular RL algorithm, most famously employed for OpenAI DotA bot) can be thought of as a cheap approximation to it. people also use something called
460.
by currymj 6y ago
i think a course like that could be really good, but I don’t think you could avoid telling people about the concept of derivatives or integrals. If you tried I think it would just make things seem more confusing and arbitrary. and you might
461.
by currymj 6y ago
teaching basic probability and statistics to everyone would be great. basic statistical education seems to me like you can get it done at middle school level -- all you need is addition and multiplication, fundamentally. i'm not an exp
462.
by currymj 6y ago
the problem is a lot of the math curriculum is designed to teach engineering computations that were only useful in the slide rule era. nobody nowadays needs to know how to hand-calculate gnarly trigonometric integrals, or solve systems usin
463.
by currymj 6y ago
i think most people would just use TF/PyTorch and ignore the specific technology on the backend. not much GPU specific stuff to learn -- very, very few deep learning people write their own CUDA code. so the question is just -- when wil
464.
by currymj 6y ago
just make sure it's NVidia. whatever graphics card you want -- all their consumer cards will work great for deep learning. make sure your motherboard and processor support whatever the newest version of PCIe is -- a major factor with d
465.
by currymj 6y ago
This feels like it would show up in a sci-fi movie from 20 years ago. I haven’t had that kind of futuristic feeling in a while. Even if this is objectively not an improvement.
466.
by currymj 6y ago
there is some truth in this. but people play and enjoy American Truck Simulator. Elite sometimes seems like Galaxy Truck Simulator. many similarities, including having to play a parking minigame before delivering your cargo.
467.
by currymj 6y ago
the original creator of Elite made a new version of the game, called Elite: Dangerous. It's the same essential game, but online and with modern, realistic graphics: https://www.youtube.com/watch?v=uoUJVRCbTxI The reall
468.
by currymj 6y ago
what i recall was that there were many pitfalls even to do something simple, and this event suggests that even experts may find it difficult to avoid mistakes. part of the challenge was just the inherent weirdness of the ethereum execution
469.
by currymj 6y ago
i tried writing some toy Ethereum smart contracts circa 2016. at that time it was immensely difficult to write them in a secure way -- even a simple "hello world" level Solidity contract could easily have exploitable bugs if you d
470.
by currymj 6y ago
i actually expect that this will work pretty well, at least for academic articles, because the machine learning engineers working on it will constantly be "dogfooding" it, as machine learning research is published mostly as PDFs.
471.
by currymj 6y ago
a lot of this was legally enforced until the 1970s. see e.g. the Racial Integrity Act in Virginia for one example. the state had a legally-codified definition of race ("one-drop rule") and put a lot of bureaucratic effort into kee
472.
by currymj 6y ago
i guarantee you there are a number of people every year who get rejected from Berkeley but accepted to Stanford (or equivalent) -- just a lot of noise in the process and after a certain point it becomes pretty arbitrary. i bet this fact is
473.
by currymj 6y ago
If you're writing code for robots in a research setting, a lot of good advice around making code easy to deploy, doing continuous integration, logging and monitoring, etc. is just not relevant. Therefore these tools are probably not wo
474.
by currymj 6y ago
this is what Lean and other theorem provers are trying to build. which would be useful whether you have a human operator generating the proofs or some kind of automated system.
475.
by currymj 6y ago
well-specified also requires that the functional relationships are actually linear, if you're using linear models, right? this is a common critique of linear SEMs -- some ML researchers refer to a broader class of "structural caus
476.
by currymj 6y ago
yeah, i've wondered about this too. the natural thing to do is to just unroll the models in time. the classic case of this is a hidden Markov Model, but you can easily have a more complicated time-indexed DAG. There are also some inter
477.
by currymj 6y ago
i get the sense that back in the 90s, statisticians must have been really rude and dismissive to Judea Pearl, and he might not have gotten over it. it seems very plausible to me that at the time, many people outside CS would have considered
478.
by currymj 6y ago
R package "daggity" is excellent. http://www.dagitty.net/
479.
by currymj 6y ago
A lot of it was accurate in between the nonsense. I was surprised to learn that there really was a Cookie Monster virus that made you type "cookie". And the hackers are accurately depicted reverse engineering some machine code as
480.
by currymj 6y ago
i mentioned this in another comment, but there are two kinds of gambler. the stereotypically "male" kind wants to take risks, win big, prove they are skilled, etc. and will tend to play games like roulette or blackjack, or poker,

More ›