gwern

33 ms·

481.
by gwern 2y ago
Who said you can't just stipulate having the same book twice or use self-reference?
482.
by gwern 2y ago
Arguably, it doesn't really make a difference, because every possible 800-page sequence exists as a pair of 400-page books from the original Library, and so on and so forth. The 800-page, 1200-page, 1600-page and every greater length L
483.
by gwern 2y ago
I think this is a little paranoid. No one is training extremely large expensive LLMs on huge datasets in the hope that a blogger will stumble across poor 1800 Elo performance and tweet about it! 'Chess' is not a standard LLM bench
484.
by gwern 2y ago
That makes sense. Sounds interesting - do you have a favored reference? Quickly looking, I did find https://gwern.net/doc/math/1974-mathai.pdf
485.
by gwern 2y ago
Hm. So you're getting the raw URL but you don't have NoScript / block JS specifically? Can you check in an incognito window and if it still happens, ablate all your extensions in a fresh profile ( https://support.mo
486.
by gwern 2y ago
(Yes. Low-discrepancy sequences are the WP link for "a special random number generator" in /unsort, and I considered that link to also cover quasi-Monte Carlo etc. I'm not immediately sure what you mean by "φ in a c
487.
by gwern 2y ago
Yes, without the swapping JS, you wouldn't get the canonical URL. But browsing Gwern.net these days without JS is pretty painful. And in this particular case, there is only one place on Gwern.net that the link exists where you could se
488.
by gwern 2y ago
I don't think that would make it any clearer. Why would 'doc.gwern.net' be more obviously just a random document than 'gwern.net/doc/www/'? Regardless, I am puzzled how OP got this URL in the first pl
489.
by gwern 2y ago
One neat thing about the AUNN idea is that when you operate at the function level, you get sort of a neural net version of lazy evaluation; in this case, because you train at arbitrary indices in arbitrary datasets you define, you can do wh
490.
by gwern 2y ago
> The link in this paragraph goes to a post on gwern website. This post contains various links, both internal and external. But I still failed to find one that supports claims about Newton's views on "progress". Could you
491.
by gwern 2y ago
I think people tend to wildly overestimate how expensive hosting a large dataset has to be, because of cloud. If you only need a few terabytes, you can rent from Hetzner for more like <$20/month. Maybe <$10 at this point, if you
492.
by gwern 2y ago
One of the things about offline imitation learning like OP or LLMs in general is that the more important the error in their world model, the faster it'll correct itself. If you think you can teleport across a river, you'll make &a
493.
by gwern 2y ago
Because it looks like a little chess king, instead of anything to do with Westinghouse or electronics?
494.
by gwern 2y ago
There are only 14k, but they are very unevenly distributed, so you might be surprised how many are left for the taking: https://gwern.net/tla You only have to go as far as 'CQK' before there is, AFAICT and based o
495.
by gwern 2y ago
Lazy people have been lazy forever. I stumbled across an example of this the other day from the 1990s, I think, and was shocked how much the student emails sounded like LLM prompts: https://www.chiark.greenend.org.uk/~martin
496.
by gwern 2y ago
Classic oldschool professor website! It's a pity that the gold leaf usage probably looks far better in person, though, compared to a digitization. Gold tends to be a material that doesn't translate onto computer screens well, like
497.
by gwern 2y ago
> There were, of course, obstacles. One of the less appealing features of Golden’s business practice was that, with rare exceptions, they offered no royalties. This arrangement nagged at Scarry, especially after his and Patricia’s son Hu
498.
by gwern 2y ago
If it's so powerful, why is Bezos also apparently now pushing for a lot of changes in the actual writing and ordered conservative writers added to the op-ed board? https://www.nytimes.com/2024/10/25/busin
499.
by gwern 2y ago
Fulltext: https://gwern.net/doc/psychology/vision/2024-xu.pdf
500.
by gwern 2y ago
"Can LLMs be Scammed? A Baseline Measurement Study" https://arxiv.org/abs/2410.13893 presumably.
501.
by gwern 2y ago
> The newspaper also published an article by two staff reporters saying that editorial page staffers had drafted an endorsement of Democratic nominee Kamala Harris over GOP nominee Donald Trump in the election. This is a bizarre way to u
502.
by gwern 2y ago
At this point, most of the stuff in a corrected edition of _Influence_ would have a bunch of asterisks in it or have to be deleted for failing to replicate or being fraudulent. I wonder how well this old social psych priming/influence
503.
by gwern 2y ago
If your timekeeper doesn't perform adequately, there's an impressively overengineered neural net project to read clocks: https://arxiv.org/abs/2111.09162 "It's About Time: Analog Clock Reading in th
504.
by gwern 2y ago
> Is there much in the way of datasets of things that are recognizable but triggering incorrect something-or-other (what is it exactly? Semantic sensations?) There are some datasets which collect 'hard' examples. One I love for
505.
by gwern 2y ago
Assuming the summary is accurate. Which I'm not sure is the case - as it happens, yesterday an actual LLM summary of OP struck me as missing important parts: https://www.reddit.com/r/Longreads/comments/1g
506.
by gwern 2y ago
It has a link to the original paper clearly visible right at the bottom where they usually are in scientific press releases like this? > The study, which was published in Science Advances, can be found at https://www.science.o
507.
by gwern 2y ago
Presumably, you're using Firefox or Safari, and so the `initial-letter` CSS property they're trying to use to make a big serif 'R' dropcap letter is breaking: https://developer.mozilla.org/en-US/docs
508.
by gwern 2y ago
Note for those suffering from deja vu: this was extensively highlighted and touted by critics back in July 2023 as "Are Large Language Models a Threat to Digital Public Goods? Evidence from Activity on Stack Overflow" https:&#x2F
509.
by gwern 2y ago
Amazing how every year since 2014 or so, someone would call DL a bubble or a grift, and yet, here we are: AI keeps getting better, while journalism & editorial standards keep getting lower. I mean, it used to be that journalists had the
510.
by gwern 2y ago
He is adapting one of Samuel Johnson's most famous quotations, about the astonishing sight of seeing a woman preaching - like a dog walking, it may not be done well, but it's astonishing to see it done at all.

More ›