currymj

71 ms·

781.
by currymj 8y ago
to my understanding, a bunch of nested if-else statements is an algorithm, and if you change all the conditions in the if statements, you get a different algorithm. so e.g. ID3 is an algorithm, designed by a human, that produces new algorit
782.
by currymj 8y ago
If you do this, using PyTorch’s new C++ frontend and API might be better; I think PyTorch is generally more popular, at least for use by individuals (maybe MXNet is better in production or something). https://pytorch.org/cpp
783.
by currymj 8y ago
But it's not really literate programming as Knuth envisioned it. A very important point was that you could write things in any order, and then use tangle to produce source code in the correct order for execution. Jupyter notebooks don&
784.
by currymj 8y ago
I actually think this is a pretty literal description of automatically learned decision trees, which are used in practice (directly or as sub parts of state-of-the-art techniques like gradient boosting). It is basically learning a series of
785.
by currymj 8y ago
There are differences in usage. For example, it is not correct usage to call someone “a French”. The adjective “transgender” seems to work similarly.
786.
by currymj 8y ago
As linguists will sometimes joke, English spelling is extremely useful, you can look at an English word and tell exactly how it was pronounced 700 years ago. Even native English speakers struggle with it. This is the so-called "curse o
787.
by currymj 8y ago
If you hire mostly men directly out of top US colleges you'll get a lot of this just by default, because that's the culture on those campuses. I get the sense that in the past, relatively fewer people came to tech through that par
788.
by currymj 8y ago
It seems like the fine for noncompliance for an individual is 238 "penalty units", which currently corresponds to nearly $50,000 (Australian), unless I misunderstand things. A $50,000 fine is quite serious even for a well-paid sof
789.
by currymj 8y ago
Neither, they mean automatic differentiation [1], which is not symbolic like SymPy or Mathematica, but also is not finite differences. Often, the interface for doing this is to just extend an existing numerical or array type and overload al
790.
by currymj 8y ago
i don’t think either is better or worse, it’s just different strategies. a probably somewhat wrong but evocative way to think about it is that what AlphaZero does is closer to “intuition”; it can look at a board and immediately tell very we
791.
by currymj 8y ago
It's a pretty common assumption that any physical process could be simulated by a Turing machine (given infinite time and space etc.). We could probably never build such a Turing machine. I didn't mean to make any stronger claims
792.
by currymj 8y ago
Why not? A human given enough time and an infinite tape can easily simulate a Turing machine. The other way is more controversial, but I don't think it's outrageous to speculate that a brain could be accurately simulated by a Turi
793.
by currymj 8y ago
it's not really mimicking how our brains work, although that was the initial inspiration, decades ago. There are still people trying to model the actual brain but they don't overlap much with the machine learning community anymore
794.
by currymj 8y ago
I don't think it would make it any easier to compare schools. I'm fairly certain a 30th percentile student at Caltech could go to Directional State University and be at the top of the equivalent class. People trying to judge stude
795.
by currymj 8y ago
i don't know about this. how much worse, really, are hanzi (which do have a loose structure to them, it's not just brute memorization) than the deranged English spelling system? if there's anyone who learned both as additiona
796.
by currymj 8y ago
this is how I understand it. but then I’ve heard statisticians describe neural networks as “nonparametric”, even though they typically have a fixed number of parameters. (millions of parameters! arguably they are the MOST parametric.)
797.
by currymj 8y ago
As far as I can tell the “free energy principle” is just asserting that the brain is approximately Bayesian and is doing some kind of variational inference, right? I’m not sure how revolutionary that is. (I’m predisposed not to like Friston
798.
by currymj 8y ago
I would like to know what statisticians mean by "nonparametric", and what machine learning people mean by "nonparametric", because they seem to be something very different.
799.
by currymj 8y ago
it's great as long as everything you need is already in their repository. if you actually have to write your own code to build things, though, you're looking at one extremely hirsute yak.
800.
by currymj 8y ago
i agree the word “AI” ought not be used to describe this. the thing that makes these sorts of problems not quite so simple is that there are fairness constraints, not just constraints on routes and schedules. there may also be a requirement
801.
by currymj 8y ago
That sounds like it’s probably right to me. But so do lots of things that turn out to be wrong. I wish we had a better grasp of what is happening, not just plausible stories. I’m already sick of doing alchemical tinkering to find a model th
802.
by currymj 8y ago
The paper “Understanding Deep Learning Requires Rethinking Generalization” was where this was first pointed out, I think. They shuffled the labels on their datasets, so there can’t possibly be anything to learn, yet got zero training loss,
803.
by currymj 8y ago
many of them are Turing-complete almost by accident, if you are willing to add enough parameters. E.g. a shallow neural network with one layer is Turing-complete. It's actually Turing-complete for rational-valued weights (countable). I
804.
by currymj 8y ago
I don't think it's intended for a general audience. It's intended mainly for people who have bought in to the idea of "effective altruism", and want a global-utility-maximizing career, and also at least somewhat bel
805.
by currymj 8y ago
This website is made by and somewhat for a subculture of people very worried about safe AGI. The idea is it's a career guide for people who want to know how to have a career that will maximize some definition of total utility (in their
806.
by currymj 8y ago
Thanks for the response, that is pleasantly surprising default behavior.
807.
by currymj 8y ago
Are Unicode characters a problem for the tools you use? For example, some languages will let you use λ instead of lambda. Or when trying to define the logistic function, I sometimes find myself writing σ(x) = 1.0 / (1.0 * exp(-x))
808.
by currymj 8y ago
It seems like if you do anything with financial data like this, people often aren’t capable of accepting that it’s just a toy example — they judge it as if it must be intended to really make money. I don’t know why.
809.
by currymj 8y ago
no, not inherently. it helps, though. note that a PhD in a computer science department is going to be much less bad than the humanities PhDs described in the article. - they are fully funded, stipends are somewhat higher (at least not total
810.
by currymj 8y ago
Yes, definitely. You can see it among the benchmarks here. https://www.pugetsystems.com/labs/hpc/RTX-2080Ti-with-NVLINK... It's the slowest tested obviously, mainly because of limited memory (so you have to p

More ›