currymj

70 ms·

871.
by currymj 8y ago
CRAN has extremely strong requirements around testing, compatibility, documentation, all aggressively enforced by the maintainers of the repository. The Julia package ecosystem is much more anarchic, like npm. You basically just have to hav
872.
by currymj 8y ago
R and Python are still (currently) nicer for stats and ML, although there's a lot of not-yet-mature Julia libraries that are extremely cool (Turing.jl for Bayesian inference in particular, and systems for autodiff). I also think Plots.
873.
by currymj 8y ago
Apart from being close to MATLAB, a goal of Julia is to make it easy to just type mathematical formulas straight off the page and into your code. 1-based indexing makes this so much easier, even if it seems somewhat depraved to a computer s
874.
by currymj 8y ago
The examples shown are all quite traditional and conservative. If you go to a lot of college towns you can get a pizza topped with taco ingredients, or fried chicken and bleu cheese, or even pesto sauce and cheese-filled tortellini.
875.
by currymj 8y ago
During training they were given access to each other’s reward functions, and the extent to which they weighted total team reward over their own was gradually increased.
876.
by currymj 8y ago
I would bet you and I evaluate the problem similarly, I think there are remote but plausible scenarios that are almost intolerably bad (and these become likely if we keep polluting). But there are left-leaning but not scientifically literat
877.
by currymj 8y ago
As far as I can tell, we are reasonably certain that the effects of climate change will be very bad. It’s hard to say exactly what will happen, though. The truly apocalyptic predictions are just not very likely. We are not going to be boile
878.
by currymj 8y ago
In some countries, especially in Europe, there are even tables ranking all journals and essentially assigning a point value. Whether you get promoted or fired depends on how high your score is.
879.
by currymj 8y ago
I think your academic career counts as experience. I would think your best bet is data science/machine learning. If you have read and mostly understood ESL i’d think you’re basically qualified. Play around with Tensorflow/PyTorch
880.
by currymj 8y ago
First, they only work on really specific problems. They aren’t just a magic tool for brute force. Current public-key crypto (both RSA and elliptic curve) happens to be one of those problems. However, there are systems where we don’t know ho
881.
by currymj 8y ago
A company called Grammofy tried to do this, it was good, but then they ran out of money. They’re back in some capacity but now tied to Spotify.
882.
by currymj 8y ago
i've read about this phenomenon before. speculation was that many/most prices are set automatically, and that these absurd prices were somehow attempts to manipulate other prices. money laundering also seems like a good explanatio
883.
by currymj 8y ago
It could make sense. If I have to work on someone else’s code I’d much rather it have a suite of unit tests than not. Having (useful, properly done) tests is good, so TDD might be useful even if the quality of the code remains the same.
884.
by currymj 8y ago
that second quote is taken out of context. The full quote is: “It’s also banning meat (but not fish) from all corporate events on the grounds, handed down by co-founder Miguel McKelvey, that “avoiding meat is one of the biggest things an in
885.
by currymj 8y ago
Part of the reason they are big in the US may be the fact that we drive so much. There are many Starbucks stores with drive-thru windows. If you want to consume a beverage while driving, a covered cup with a straw is much easier. (One could
886.
by currymj 8y ago
There’s an evolving social norm against this, I think. Increasingly, I’ve noticed the people who trawl through years-old posts looking for controversy are seen as deranged.
887.
by currymj 8y ago
I agree strongly. His description of what cyberspace looks like is absolutely perfect: "Cyberspace. A consensual hallucination experienced daily by billions of legitimate operators, in every nation, by children being taught mathematica
888.
by currymj 8y ago
The book "Void Star" by Zachary Mason was published last year. A high concept pitch for it is "Neuromancer, but all the 80s stuff is replaced with 2018 stuff". I hope the author wouldn't be offended by that -- the b
889.
by currymj 8y ago
i didn't remember, but searching around it looks like it was something he mentioned in an interview with Bryan A. Garner.
890.
by currymj 8y ago
this is similar to an exercise David Foster Wallace would do — take a longish piece of prose, read it once, then try to replicate it exactly without looking. Assuming it’s long enough that you can’t hold it in working memory, you’ll end up
891.
by currymj 8y ago
if your only goal is to get your work published in something that is technically a journal, this is very easy even if the research is very bad, and then it will likely end up on PubMed. fundamentally, it is just a specialized search engine,
892.
by currymj 8y ago
Here, https://www.ncbi.nlm.nih.gov/pmc/about/disclaimer/ . That gives a pretty clear explanation. Quote: "The presence of an article in PMC does not reflect an endorsement of, or concurrence with, the co
893.
by currymj 8y ago
it's important to note that the nih.gov domain doesn't mean this is endorsed in any way by the NIH. (In case you were very shocked, like I was.) It's just a link to PubMed, which is an aggregator of journal articles hosted by
894.
by currymj 8y ago
AMD is scrambling to build a CUDA-esque thing called ROCm and get various frameworks including TF ported over, but it’s not even in “early adopter” state yet, so i doubt anyone is using it. However they seem to be making progress.
895.
by currymj 8y ago
Yeah, I see what you're getting at now. And that is an excellent diagram. I've never actually used BUGS or JAGS for anything but I would consider at least BUGS to truly be legacy software. To me the word "legacy" means y
896.
by currymj 8y ago
It's definitely not a legacy framework, it's very actively developed and improved. A lot of very useful statistical models can't benefit from the GPU, and for them I think Stan is the better tool. There's a reason it&#x2
897.
by currymj 8y ago
Stan is probably the most mature, stable PPL, and it's an extremely popular tool, although it isn't really deep-learning-adjacent so gets much less hype. It's generally used more for modeling and prediction than for creating
898.
by currymj 8y ago
i don't think it's nonsense exactly. i think it merits some funding, and we'll probably be glad we did it even if it turns out there never was a risk of an unfriendly superintelligence. but I have misgivings. in particular,
899.
by currymj 8y ago
the author (Tyler Cowen) regularly talks about how Singapore has a particularly effective and well-run government.
900.
by currymj 8y ago
if you are an engineer working for one of these companies, you should get a different job as soon as possible.

More ›