currymj

26 ms·

331.
by currymj 4y ago
a lot of the fear is not so much not what the law says, but how a local school administration or local judge decides to interpret it after arbitrary complaints by parents. even if you’re eventually ruled to comply with the law it could make
332.
by currymj 4y ago
in the humanities there is a glut of excellent scholars who would be happy to have a non-tenured professorship which is nevertheless a real job with decent pay, benefits, and working conditions. in applied technical fields it will be really
333.
by currymj 4y ago
I think for the application here (analogous to autograd on neural networks) you know each individual building block of the function (a well-defined neural network layer) so can be sure such things will not happen.
334.
by currymj 4y ago
they don't pay the grad student $46k a year. they pay them $24k a year (or more likely $20k for 9 months), and then the university makes up a number for "tuition" and takes an extra $15k arbitrarily.
335.
by currymj 4y ago
From the perspective of a grad student, you're getting paid $20k/year in a HCOL area, the grant is getting charged like $40k a year (somewhere around $20k of "tuition remission" charged on fake tuition that nobody actual
336.
by currymj 4y ago
this has always struck me as a weird argument. one could make a very rough comparison to the energy required to run a gaming PC for an hour, and the energy (inference + amortizing the training run) to run a conversation with ChatGPT for an
337.
by currymj 4y ago
there’s tons of research in that general direction. not sure if any is practically better yet but it does “work”.
338.
by currymj 4y ago
there's active work on watermarking LM outputs. here's one example: https://huggingface.co/spaces/tomg-group-umd/lm-watermarking
339.
by currymj 4y ago
there's been steady progress on language models for the past 6-7 years, getting somewhat faster in the last 2 years, in part due to fundamental advances, in part due to (mainly) OpenAI being a little more product focused. it's tru
340.
by currymj 4y ago
they didn't initially. their first mobile game, Super Mario Run, cost $10 and had none of the usual manipulative mobile game elements. People did not buy it so they switched to the same F2P nonsense as everyone else for future games.
341.
by currymj 4y ago
The recent Dune movie is excellent and is rated PG-13. Same with Gattaca.
342.
by currymj 4y ago
The odd thing is, the motivation for the research was not really to build this stuff as a product. If you're going to do machine learning dealing with images, music, or text, a completely natural thing one might want to do for many rea
343.
by currymj 4y ago
Supreme Court ruled that public sector employees can't be required to pay union fees as a condition of employment, on First Amendment/free association grounds.
344.
by currymj 4y ago
in my experience it happens pretty regularly if you ask one of these things to generate code (it will often come up with plausible library functions that don't exist), or to generate citations (comes up with plausible articles that don
345.
by currymj 4y ago
just skip ahead!
346.
by currymj 5y ago
congrats on the launch! We talked briefly a while back, and I've been checking in on your page every so often to see when things would finally get rolling. The world needs more mechanism design. Recently in school I've been thinki
347.
by currymj 5y ago
In ML at least, the arXiv version is the “canonical” one. This is because the conferences have onerous page limits, so the official conference version is usually mangled with lots of cut content. The arXiv will have content restored and be
348.
by currymj 5y ago
thanks, this is going to be very helpful for me. i guess it’s kind of like that old piece of advice, if you want some free Linux tech support, just post “Linux can’t do this but Windows can” :)
349.
by currymj 5y ago
PyTorch is working on catching up — I think they’ve already got some kind of “vmap” style function transformations in beta. And I’m sure they’ll figure out good higher order derivatives too. That’s like 90% of what people want out of Jax, s
350.
by currymj 5y ago
the price was reasonable for a good piece of software, but for the way people want to do ML research now (launch tons of parallel jobs on lots of cloud machines) the price, and more importantly, the burden of managing licenses, was prohibit
351.
by currymj 5y ago
you could install that license on one physical machine. so say you want to use a Colab or cloud GPU or something, or your university compute cluster, it wouldn’t work.
352.
by currymj 5y ago
DeepMind’s OpenSpiel repository on Github might be a good place to start, they provide infrastructure for a lot of games including poker, and example implementations of algorithms like CFR.
353.
by currymj 5y ago
This interactive is an interesting demonstration, but if you want to just read the article, it's here (paywalled) https://www.ft.com/content/c0b03d1d-f72f-48a8-b342-b4a926109...
354.
by currymj 5y ago
I recently ran into the concept of a "variational inequality". A variational inequality defines the optimality conditions for a problem in terms of some F. It basically says that x is the optimal solution to the problem if "f
355.
by currymj 5y ago
The US medical residency was mostly created by William Halsted, a very influential surgeon. He was partly responsible for this culture of sleep deprivation, insisting that residents be on-call 24/7. He also was a serious cocaine addict
356.
by currymj 5y ago
America Magazine is run by Jesuit priests, among the most left-leaning of Catholic organizations, they're editorially both pro-labor and pro-child support.
357.
by currymj 5y ago
i think some descendant of cryptocurrency technologies has a good chance of being in use in 100 years, it's very cool stuff. will it be Bitcoin? i don't know, I think it's likely something better will come along but there are
358.
by currymj 6y ago
it is a particular part of the Python 3 standard library. it is one extra line of code to automatically memoize a recursive function. you just do: from functools import lru_cache @lru_cache(maxsize=None) def brute_recursi
359.
by currymj 6y ago
San Francisco did a reform to the math curriculum that angered a lot of people. However, you can see how the courses in SF work here: https://www.sfusdmath.org/high-school-pathways.html So after the reform, despite all the
360.
by currymj 6y ago
how much are these numbers skewed by academic monographs? these are books that are basically going to be sold only to university libraries, which might explain why they sell <500 copies. there are many such books where no author ever exp

More ›