Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
arugulum
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
11 ms
·
61.
▲
by
arugulum
6y ago
>Any place that executes people for minor crimes or non-crimes is pretty awful on human rights issues. I am going to press you on this. Which non-crimes lead to capital punishment? And on what basis do you get to decide whether a crime i
62.
▲
by
arugulum
6y ago
It is hard to take a comment like this in good faith. You make multiple references to Singapore being a dystopia with zero justification. You make several strongly negative analogies of Singapore ("deal with the devil", "cost
63.
▲
by
arugulum
6y ago
I'm struggling to understand on what basis you're saying "almost nothing is oriented toward children". For instance, how many schools did you visit during your two weeks there? Or what are you considering to be "ori
64.
▲
by
arugulum
6y ago
By the Python standard library already does contain things that don't follow best practice. See: logging doesn't abide by PEP 8.
65.
▲
by
arugulum
6y ago
"You have a model evaluation problem. You decide to use Frechet Inception Distance. Now you have two model evaluation problems."
66.
▲
by
arugulum
6y ago
> Internships that want to be converted to FTE needs to try the H1B lottery also. This is not true in general, assuming you're a student (since that's the usual situation for an internship). You can rely on CPT during school (w
67.
▲
by
arugulum
6y ago
H-1B is dual intent.
68.
▲
by
arugulum
6y ago
Apologies for misusing the term "immigration" to refer to the broader topic, but in any case H-1B is a dual-intent visa and is still the primary work-related route on the way to applying for a green card.
69.
▲
by
arugulum
6y ago
My understanding is that O-1 visas require very specifically tailored applications to qualify. (I too have heard stories of how moderate achievements are twisted/massaged to be pitched for the O-1 application, and worked.) I know sever
70.
▲
by
arugulum
6y ago
I will repost my quick immigration quiz, for Americans/others who may be unfamiliar with what the typical visa process is for an immigrant: 1. An international student graduates at the top of their class in Stanford in CS, and goes to
71.
▲
by
arugulum
7y ago
If I remember correctly, this wasn't really a "mathematics" mistake. The formula simply wasn't applied to the right number of cells (they didn't apply the cell-selection far enough). Although that's beside the
72.
▲
by
arugulum
7y ago
It is the primary path for most people to get a work visa. No, there isn't some other work visa that is available and viable that people are likely to obtain. Only people in very specific and special situations get something else. Th
73.
▲
by
arugulum
7y ago
Exactly. I have no doubt that H-1B has been and continues to be abuse. But it's also literally the only viable work visa in the majority of cases, so painting with a broad paintbrush about the exploitation of H-1Bs is a bad idea.
74.
▲
by
arugulum
7y ago
Reposting my quick immigration quiz: 1. An international student graduates at the top of their class in Stanford in CS, and goes to work for a tech company. What visa do they use? The correct answer is F1 (OPT) for one year, with OPT STEM e
75.
▲
by
arugulum
7y ago
Because this has become a pattern in OpenAI's MO. Take a very impressive research achievement (large LSTM on byte-level language modeling/GPT-2). present it in a hyperbolic manner ("we've discovered a single neuron that
76.
▲
by
arugulum
7y ago
I will add further to this point. There will be bad model code in both PyTorch and TensorFlow. The difference is that, bad PyTorch code reads like bad Python code, and I've accumulated a lot of experience reasoning through bad Python c
77.
▲
by
arugulum
7y ago
There has basically only ever been one major API shift, which was the shift away from Variables. And, granted, that happened from 0.3->0.4, about a year after the initial release. Other than that, I've had next to no issues, and the
78.
▲
by
arugulum
7y ago
Granted that PyTorch and TensorFlow both heavily use the same CUDA/cuDNN components under the hood (with TF also having a billion other non-deep learning-centric components included), I think one of the primary reasons that PyTorch is
79.
▲
by
arugulum
7y ago
IAGO Good my lord, pardon me: Though I am bound to every act of duty, I am not bound to that all slaves are free to. Utter my thoughts? Why, say they are vile and false; As where's that palace whereinto foul
80.
▲
by
arugulum
7y ago
My own experience with dataclasses vs attrs: `dataclass` primarily just provides a faster syntax for defining a simple struct-like class. For that alone I've found it tremendously useful. At the same time, sometimes you do need more fu
81.
▲
by
arugulum
7y ago
Here's a quick immigration quiz: - An international student graduates at the top of their class in Stanford in CS, and goes to work for a tech company. What visa do they use? The correct answer is F1 (OPT) for one year, with OPT STEM e
82.
▲
by
arugulum
7y ago
The key part analogy isn't bankruptcy so much as interest. You incur technical debt and you pay it off later in interest (by spending time to clean up the hacky code). If you don't pay off the interest, you end up incurring even m
83.
▲
by
arugulum
7y ago
Balance of power. You can draw an analogy to price elasticity in economics (more directly, the cost of switching). Because of identity theft, the expected cost of business between you and the bank has gone up. Who has an easier time switchi
84.
▲
by
arugulum
7y ago
I would love to read a deep dive into the history of (semi) modern science. In particular, one that would avoid pop-science analogies and that is unafraid of technical detail. Ideally one that is more focused on the intricacies and lineage
85.
▲
by
arugulum
7y ago
>If the death note can manipulate odds to an arbitrary degree, you're thinking way too small right now. Right, what I feel might be the case is that the targeted output needs to at least be verifiable (hence see my second proposal).
86.
▲
by
arugulum
7y ago
Agreed, but there is still so much you can do. You can make dictators liberalize their countries. You can make billionaires donate away their wealth. And so on.
87.
▲
by
arugulum
7y ago
>Light is powerhungry and arrogant, but he's also very moralistic. A character that proritizes fear over accuracy would be a different character than Light. He wants to build a better, even perfect, world. And for that, he must have
88.
▲
by
arugulum
7y ago
>I think the most under-utilized ability of the Death Note in the series is the ability to specify manner of death. Absolutely. This ability is so significant that it completely overshadows the ability to kill people. If anything, the de
89.
▲
by
arugulum
7y ago
>I'm pretty sure this would be interpreted as "The person has to flip coins until they have a 100 streak of only heads". I disagree. For example, Light has people die of getting run over a by truck. But the Death note does
90.
▲
by
arugulum
7y ago
Well we know the reason why Light says he can't do that is that "the person wouldn't know the name". But you're right in actuality that would allow for "person guesses the name, and is correct". And I thin
More ›