HNHacker News
TopNewBestAskShowJobs

C-x_C-f

326 karma · joined April 20, 2021

submissionscomments
C-x_C-f··on The tragedy of the commons, AI edition
Yes, although they don't really obviate the issues raised by GP (costs, length, complexity).

If anything, some of the largest civil law countries in Europe, like Germany and Italy, can arguably be considered worse than common law countries in all of these respects. Whether that's incidental or due to civil law, I cannot say.

C-x_C-f··on Don't use your phone while you poop
https://www.youtube.com/watch?v=DJklHwoYgBQ
C-x_C-f··on How to Exist
I might just be too autistic for this but I don't get why the article uses the word "existing" to refer to a conflation of contentedness/meditation/stasis/passivity/what have you.

Am I not existing when I'm worrying about finances or fidgeting with my hair?

I don't really see why "existing" should be conceived in such a subtractive fashion.

I'm not dismissing the value of contentedness/meditation/stasis/passivity when achievable (which, as the author acknowledges, these things sometimes aren't). But I am slightly miffed by the implication that any failure to reach such state is an inability to attain something as fundamental as "existence".

(end rant)

C-x_C-f··on The footprints of every building in NYC
> Censored is a very strong word

I believe GP is using the stats/data science meaning of the word "censor" which is a neutral term simply indicating missing/obscured data (without implication of intent) [0]

[0] https://en.wiktionary.org/wiki/censor#Verb

C-x_C-f··on A Mathematical Tribute to the Soccer Ball
You can even find the exact number of pentagons (or hexagons, or edges, or vertices) using the equation

V - E + F = 2

where V, E, F are the number of vertices, edges, and faces, respectively.

This holds for any polyhedron (and other shapes have similar equations possibly with a different right hand side) and the left hand side is called the Euler characteristic of the soccer ball (or any polyhedron).

Spoiler warning: the Wikipedia article for the Euler characteristic [0] has a worked out example specifically for the soccer ball.

[0] https://en.wikipedia.org/wiki/Euler_characteristic

C-x_C-f··on My Mathematical Regression
That's a nice way to put it.

FWIW the generalization of binomial coefficient which allows you to express an n-dimensional solution is called a multinomial coefficient [0]. So in a 3d 20x20x20 box we would have (60 multichoose 20, 20, 20) paths.

Also, the wiki article doesn't mention this but the growth rate of (n multichoose k1, k2, ..., km) as we increase n but fix the ratios p1 = k1 / n, ..., pm = km / n is precisely the Shannon entropy of the categorical distribution with probabilities p1, ..., pm . The wiki article for entropy [1] states the result for the binomial coefficient, which can be written as (n choose k) = (n multichoose k, (n - k)) .

Actually a lot of basic information-theoretic results about entropy and related quantities (e.g. the properties of the Boltzmann distribution/softmax function) can be derived from similar discrete counting problems after taking a large-n limit. I don't have links at the ready but I might edit this comment if I remember places which explain this stuff.

[0] https://en.wikipedia.org/wiki/Multinomial_theorem#Number_of_...

[1] https://en.wikipedia.org/wiki/Entropy_(information_theory)#A...

C-x_C-f··on Building Pi with Pi
> All these models see a local failure and try to locally defend against it. As maintainers we have to keep pulling the conversation back to the global invariant, which is harder than it should be, and it’s laborious.

This has been by far the biggest and costliest failure mode I've experienced using these tools. I've tried to mitigate it in more ways than I can count but it almost feels structurally impossible for LLMs to get this right.

C-x_C-f··on ISBN Visualization
That's funny, I came out of it with the opposite impression.

In the section for my native tongue, I zoomed in a few times, here and there, and I only once did I stumble upon an author I knew.

To me, writing a book feels like such a monumental endeavor that I find it hard to fathom the amount of collective effort that it took to write all this, especially considering how most of these works are almost forgotten by now (something something power law).

C-x_C-f··on Origami Patterns Solve a Major Physics Riddle
The author mentioned (Galashin) has some fun software to play with these combinatorial structure.

I went to a talk of his that was "interactive" in that to follow along you could play with an interactive widget, but I can't seem to find it right now. This one about plabic tilings (objects mentioned in the article) is fun too: https://www.math.ucla.edu/~galashin/plabic.html

C-x_C-f··on Borehole Oscillators
Funnily enough I went into it with a background in math and was surprised about one specific claim that I couldn't quite understand, and it turns out it was subtly incorrect in such a way that it actually adds an interesting twist to the story (Greg Egan acknowledged it). I can't quite find the web page with the discussion (ETA: found it, it's the addendum at the end of the FAQ about the book [0]) but it's about <spoilers>the Garden of Eden configuration of the automaton.</spoilers>

ETA: I realize this sounds nitpicky and stickler-y so I just want to point out that I loved the book (and Greg Egan's work in general) and figuring out the automaton stuff was genuinely some of the most fun I've had out of a book.

[0] https://www.gregegan.net/PERMUTATION/FAQ/FAQ.html

C-x_C-f··on Borehole Oscillators
Spring motion is the motion of systems where the force is proportional to the distance.

Many interesting systems (like springs) are near equilibrium, which means that the potential energy is at a local minimum. A spring is an example, but also a pendulum.

When the potential is at a local minimum, its gradient is zero. So if you Taylor expand it you only get second-order contributions. For a spring, the potential energy looks like V(x) = V(0) + k * x * 2 where x is the displacement and k is a constant.

Differentiating, you get harmonic motion: F(x) = k * x

Broadly speaking, this applies to all systems near equilibrium, simply from Taylor expanding the energy. And it's not only in classical mechanics, but in all branches of physics. Sydney Coleman [0] is often quoted as saying something like "QFT is simple harmonic motion taken to increasing levels of abstraction." [1]

[0] https://en.wikipedia.org/wiki/Sidney_Coleman

[1] https://physics.stackexchange.com/questions/355487/qft-is-si...

C-x_C-f··on Restriction on Entry of Certain Nonimmigrant Workers
> America got to the moon without H1-Bs.

https://en.wikipedia.org/wiki/Operation_Paperclip

From the linked article:

> The operation played a crucial role in the establishment of NASA and the success of the Apollo missions to the Moon.

C-x_C-f··on Restriction on Entry of Certain Nonimmigrant Workers
The relevant variable here is the increase in jobs. The crisis in my example is just some exogenous variable. I'm genuinely trying to understand your point. In my view, if one cared more about difference in relative increases inter-group rather than absolute differences intra-group, then scenario 1 would be preferable.
C-x_C-f··on Restriction on Entry of Certain Nonimmigrant Workers
I did considered that but then went with Hanlon's razor [0]

[0] https://en.wikipedia.org/wiki/Hanlon%27s_razor

C-x_C-f··on Restriction on Entry of Certain Nonimmigrant Workers
Do you think that the employee pool overall would be more or less talented without the H-1B program? What about the tails of the distributions?
C-x_C-f··on Restriction on Entry of Certain Nonimmigrant Workers
I still don't see what the issue is.

Consider two scenarios:

1. H-1Bs get massively reduced over the next five years. Tech sector crisis ensues due to market shocks. Foreign workers decrease. American workers go up by 25% (~2.5M).

2. H-1B numbers stay as is. Tech sector relatively unperturbed. Foreign workers increase by 50% (~1.3M). American workers increase by 30% (~3M).

Wouldn't you argue that 2 is preferable to 1?

(I'm not saying that a crisis will happen if H-1Bs end, I'm just presenting two scenarios with different relative increases that I believe prove my point)

C-x_C-f··on Restriction on Entry of Certain Nonimmigrant Workers
How are H-1Bs comparable to nepotism?
C-x_C-f··on Why language models hallucinate
> A language model is made to predict language

<pedantry>Isn't a language model made to predict the next token in a series, which just so happens to be good for predicting not only natural languages, but also formal ones (code and math)?</pedantry>

Also, similar to what nelox said, as long as language (or sequences of tokens or what have you) can be "about" something (whatever that means), then it's possible that LLMs are encoding information about that "something". I'm being deliberately vague because I think that trying to be precise (by e.g. referring to latent spaces and so on) makes it sound like we've figured something out when in reality we haven't even found the right words to ask the questions.

C-x_C-f··on Lisp from Nothing, Second Edition
I love Lisp (I'm an Emacs user and often write in Racket for personal projects) but the one thing I never understood about the Lisp community is the emphasis placed on metacircular evaluators.

I sure find them beautiful and all, but why do they take center stage so often? Beside the aesthetics and instructional value, I don't get the appeal. Also I feel that a bunch of the heavy lifting behind metacircular evaluators is actually done by the Polish notation syntax as well as the actual implementation, and these concepts don't get nearly as much love.

Any Lisper who can illuminate me?

C-x_C-f··on Gaussian Processes for Machine Learning (2006) [pdf]
These are very cool, thanks. Do you know what kind of jobs are more likely to require Gaussian process expertise? I have experience in using GP for surrogate modeling and will be on the job market soon.

Also a resource I enjoyed is the book by Bobby Gramacy [0] which, among other things, spends a good bit on local GP approximation [1] (and has fun exercises).

[0] https://bobby.gramacy.com/surrogates/surrogates.pdf

[1] https://arxiv.org/abs/1303.0383

C-x_C-f··on Show HN: Semantic Calculator (king-man+woman=?)
I don't want to dump too many but I found

   chess - checkers = wormseed mustard (63%)
pretty funny and very hard to understand. All the other options are hyperspecific grasslike plants like meadow salsify.
C-x_C-f··on Coffee Too Weak? Try This
A similar result was found by J. Gagné [0] five years ago, though I don't think he publishes his results outside of his blog (which is well worth a read by the way).

[0] https://coffeeadastra.com/2020/05/23/the-physics-of-kettle-s...

C-x_C-f··on Automating Interactive Fiction Logic Generation with LLMs in Emacs
How did you get started? Was it mostly plug-and-play or was some nontrivial hacking involved? I use emacs and I normally wouldn't mind shaving a yak or two, but right now I'm swamped with work and I'm kinda scared of getting sucked into a rabbit hole.
C-x_C-f··on Matrix Calculus (For Machine Learning and Beyond)
> Also, don't forget the Jacobian and gradient aren't the same thing!

Every gradient is a Jacobian but not every Jacobian is a gradient.

If you have a map f from R^n to R^m then the Jacobian at a point x is an m x n matrix which linearly approximates f at x. If m = 1 (namely if f is a scalar function) then the Jacobian is exactly the gradient.

If you already know about gradients (e.g. from physics or ML) and can't quite wrap your head around the Jacobian, the following might help (it's how I first got to understand Jacobians better):

1. write your function f from R^n to R^m as m scalar functions f_1, ..., f_m, namely f(x) = (f_1(x), ..., f_m(x))

2. take the gradient of f_i for each i

3. make an m x n matrix where the i-th row is the gradient of f_i

The matrix you build in step 3 is precisely the Jacobian. This is obvious if you know the definition and it's not a mathematically remarkable fact but for me at least it was useful to demystify the whole thing.

C-x_C-f··on Percolation Theory [pdf]
Ah but that's a bit different. The giant component doesn't connect all the neurons, only a fraction. The wiki page doesn't say this but if you have c * n / 2 edges then the fraction of neurons in the giant component is 1 + W(-c * exp(-c))/c where W is the Lambert W function [0], also called the product logarithm. As c tends to infinity this fraction tends to 1 but it's never 1.

[0] https://en.wikipedia.org/wiki/Lambert_W_function

C-x_C-f··on Percolation Theory [pdf]
What do you mean by well-connected topology? If you mean that you can reach every neuron from any neuron then the number of connections you need is asymptotically n log n / 2 (not up to a constant factor or anything, just n log n / 2 on the nose, it's a sharp threshold), see [0]. In general when percolation is done on just n nodes without extra structure, it's called the Erdős–Rényi model [0], and most mathematicians even just call this "the" random graph model.

[0] https://en.wikipedia.org/wiki/Erd%C5%91s%E2%80%93R%C3%A9nyi_....

C-x_C-f··on Kill your Feeds – Stop letting algorithms dictate what you think
I've mentioned this in another comment but the Freetube [0] client for YouTube has settings that let you hide pretty much every distraction apart from the video. You can even exclude videos in your search result if they contain a set phrase (great for avoiding political bait). I know there are extensions that do all these things but I find this to be a nice all-in-one solution, and the UI is more responsive too. (It does suffer from expiring sessions and the like though.)

[0] https://freetubeapp.io/

C-x_C-f··on Kill your Feeds – Stop letting algorithms dictate what you think
In the spirit of the video (and the article), I'm watching this on Freetube [0] while hiding the comments, recommended videos tab, and pretty much everything else apart from the video description.

[0] https://freetubeapp.io/

C-x_C-f··on Softmax forever, or why I like softmax
I agree that the choice of softmax is arbitrary; but if I may be nitpicky, the average weight and the temperature determine one another (the average weight is the derivative of the log of the partition function with respect to the inverse temperature). I think the arbitrariness comes more from choosing logits as a weight in the first place.
C-x_C-f··on Softmax forever, or why I like softmax
No need to introduce the concept of energy. It's a "natural" probability measure on any space where the outcomes have some weight. In particular, it's the measure that maximizes entropy while fixing the average weight. Of course it's contentious if this is really "natural," and what that even means. Some hardcore proponents like Jaynes argue along the lines of epistemic humility but for applications it really just boils down to it being a simple and effective choice.
Page 1 of 5Next →