nybsjytm

8 ms·

91.
by nybsjytm 3y ago
> Brute searching all possible mathematical constructs, theorems, etc. to see which one fits the problem would probably take you practiacally an infinite amount of time. That's not the kind of search which is being done. Read this p
92.
by nybsjytm 3y ago
The search space here is not so large. Read this paper, a key part of the present work: https://doi.org/10.1023/A:1006171315513
93.
by nybsjytm 3y ago
This is at almost the polar opposite end of the spectrum from "AGI," it's centered on brute search.
94.
by nybsjytm 3y ago
Thanks for the correction, it looks like you're right.
95.
by nybsjytm 3y ago
As said by others, that is an absolutely commonplace saying, albeit usually having nothing whatsoever to do with AI. See for instance https://terrytao.wordpress.com/career-advice/advice-on-mathe... Also if you read abo
96.
by nybsjytm 3y ago
If I read their paper right, this is legit work (much more legit than DeepMind's AI math paper last month falsely advertised as solving an open math research problem) but it's still pretty striking how far away the structure of it
97.
by nybsjytm 3y ago
No, since there isn't a differentiable structure. And even the use of "embedding" doesn't really match the way mathematicians would use the word.
98.
by nybsjytm 3y ago
>For instance, consider a unit cube in n dimensions. The volume of this cube (1 unit on each side) remains constant, but the distance from the center to a corner increases with the square root of the number of dimensions. This implies th
99.
by nybsjytm 3y ago
Yes, but the distribution doesn't have to be uniform. It can be almost anything.
100.
by nybsjytm 3y ago
It's an example of concentration of measure: https://en.wikipedia.org/wiki/Concentration_of_measure It's easy to verify this for yourself in a simple example in something like python. Draw from a multivariate
101.
by nybsjytm 3y ago
Based on the first three figures, it seems like the author isn't familiar at all with probability in high-dimensional spaces, where these phenomena are exactly what you'd expect. Because of that, the figure with intersecting spher
102.
by nybsjytm 3y ago
Rigor here doesn't have to mean mathematical rigor, it includes qualitative rigor. It's unrigorous to include meaningless comparisons to prior work (which is a credible issue the reviewers raised in this case) and it's also p
103.
by nybsjytm 3y ago
I don't consider clearly stating your model and meaningfully comparing it to prior work and other models (seemingly the main issues here) to be analogous to a proper haircut or a shibboleth. Actually I think it's a strange compari
104.
by nybsjytm 3y ago
Even if you were to take the extreme position that influence or citation counts are all that matter, the problem is that 'future influence' is hard if not impossible to judge well in the present. (Of course it's easy to point
105.
by nybsjytm 3y ago
Quality in the sense I meant it (cogency and intellectual depth/rigor) should certainly be correlated with importance and influence!
106.
by nybsjytm 3y ago
I guess your comment is against the restrictive and rigid idea that peer review should be about making research papers more intellectually rigorous?
107.
by nybsjytm 3y ago
I tried asking on another thread what Goodfellow rage he's referring to since all I could find was this: https://twitter.com/goodfellow_ian/status/113352818965167718... If so, frankly I think it makes Mikolov
108.
by nybsjytm 3y ago
I think the reviewers did a good job; the reviews are pretty reasonable. Reviews are supposed to be about the quality of a paper, not how influential they might be in the future! And not all influential papers are actually very good.
109.
by nybsjytm 3y ago
I think that still counts as hard-wired, just a general hard-wiring! Whatever you call it, it's certainly not as well known as it should be in the word2vec discourse.
110.
by nybsjytm 3y ago
I'm assuming this is the example of "Ian Goodfellow raging about it on Twitter"? https://twitter.com/goodfellow_ian/status/113352818965167718...
111.
by nybsjytm 3y ago
The found function is in here: https://github.com/google-deepmind/funsearch/blob/main/cap_s... . I'm not very familiar with genetic algorithms but it's pretty hard for me to imagine that they co
112.
by nybsjytm 3y ago
> An LLM can discover a new solution in high dimensional geometry that hasn’t advanced in 20 years!? That goes way beyond glueing little bits of plagiarized training data together in a plausible way. The first advance in 20 years was by
113.
by nybsjytm 3y ago
Some important context: the discovery is that a certain number in combinatorics is now known to be between 2.2202 and 2.756, not just between 2.218 and 2.756 as discovered last year. The improvement is by finding some particular sequences o