r-zip

10 ms·

61.
by r-zip 5y ago
Yes, Guile Emacs got abandoned, at least the last time I checked.
62.
by r-zip 5y ago
As someone who uses Emacs and likes writing lisp, 100% agree. elisp is without a doubt the worst lisp still in common use.
63.
by r-zip 5y ago
I think you mean punching up, or "isn't seen as punching down."
64.
by r-zip 5y ago
I think it definitely depends on where you are and what you're doing. In 3 of my 4 years of professional Python experience, we have run up against scaling, maintenance, and parallelism issues constantly.
65.
by r-zip 5y ago
That's a fine attitude for applied ML, but aren't you even a little bit interested in why these things work so well? It's not obvious.
66.
by r-zip 5y ago
I'd recommend anyone interested in the NTK and its limitations to check out Greg Yang's work [1] and the recent book by Roberts, Yaida, and Hanin [2]. There's clearly a lot more work to be done on neural network scaling limit
67.
by r-zip 5y ago
I disagree. Python's popularity is actually kind of tragic. The package ecosystem leads many to choose Python, which has both maintenance and performance scalability issues (not to mention the sorry state of packaging, distribution, an
68.
by r-zip 5y ago
To me, taking "principled approach" means you understand and can justify the eventual outcome of the approach, or at least guarantee that the outcome satisfies some constraints. How would you justify the number of channels in each
69.
by r-zip 5y ago
I think you have a different definition of "principled" from most people.
70.
by r-zip 5y ago
What's your point? It's not clear at all from this post.
71.
by r-zip 5y ago
Has this been implemented? What kinds of hashing functions are you talking about? How would you guarantee the same hash for all the augmentations? It seems like the approach you describe just moves the complexity of the task solved by neura
72.
by r-zip 5y ago
> And in fact I've never stated that the "model" contained the whole source dataset. Apologies for seeming rude, but I feel that the abstract is disingenuous (I'm assuming you're the author of the article). The a
73.
by r-zip 5y ago
The last iteration (or epoch) of SGD is not shipped with the trained model. The point just does not stand. There are other (better) arguments for why such models are derivative works. Basically the argument starts with a claim (you can reco
74.
by r-zip 5y ago
Right. While I appreciate the author's skepticism and diction (there is a lot of misleading terminology thrown around by the ML community), his points don't land. In particular, he argues that there's no learning going on, bu
75.
by r-zip 5y ago
Notebooks aren't great for building a product, either.
76.
by r-zip 5y ago
It is very sloppy when you come from say, analysis, where things are precisely defined, especially in more applied statistics. In my experience, any more applied than probability theory itself and you get omissions of very important details
77.
by r-zip 5y ago
I'm not sure, but this may be something to bring up with a psychologist/psychiatrist/therapist. If your memories are "overly general," then it may be linked to something else going on: https://en.wikipedi
78.
by r-zip 5y ago
Wow, this looks amazing. There has been a lot of development in this area recently, but without expertise in various areas (functional analysis, statistical physics, manifold theory), reading the papers themselves is difficult if not imposs
79.
by r-zip 5y ago
Your project looks super interesting. I unfortunately haven't been able to get it to work with native-comp.
80.
by r-zip 5y ago
Thank you for bringing sanity to the comment section. I'm getting really weary of all the "AI" snake oil (I'm a data scientist). There's no way that this will generalize to: 1. materials not in the training set 2. s
81.
by r-zip 5y ago
Strongly disagree. The existence of smartphones has accelerated the development of social media, which negatively impacts people's mental health and children's attention spans. Their popular adoption has also driven the growth of
82.
by r-zip 5y ago
In addition to what the other commenter said, Tile products are very flaky. The price difference doesn't matter to me, since Tile's products just don't work.
83.
by r-zip 5y ago
I think the commenter's point was that genetically, we were just as capable (if not moreso) before modern times. Of course people's lives and outcomes have improved; that's not in question.
84.
by r-zip 6y ago
Because by leaving on that journey you would be effectively dead to everyone you knew?
85.
by r-zip 6y ago
He's saying if you wiped out human civilization it would take O(10k) years to rebuild, because knowledge/culture isn't stored in the genes.
86.
by r-zip 6y ago
I think it's important to note that the first graphic you posted doesn't include longitudinal data (i.e., doesn't correct for the Flynn effect). Compare to: https://www.ncbi.nlm.nih.gov/pmc/articles/
87.
by r-zip 6y ago
Since the names have to be dotted anyway, why not .var_name? I feel like this implementation is seriously half-baked.
88.
by r-zip 6y ago
I think pattern matching is super powerful, reduces boilerplate, and is not that hard to understand if implemented well. I also think those who complain about the decline of Python's beginner-friendliness need to take a step back and u
89.
by r-zip 6y ago
Under certain technical conditions, differentiation under the integral sign also works for Riemann integration (see Marsden & Hoffman). There's no need to develop Lebesgue theory to demonstrate this technique in a calculus course,
90.
by r-zip 6y ago
VSCode isn't as flexible at all though. You can't even create a keybinding that executes multiple commands (macro). There aren't hooks. You can't just write a function in a file and execute it. There isn't the same

More ›