gwern

66 ms·

931.
by gwern 3y ago
No, what it said was that it didn't find evidence he cheated. A failure to reject the null does not mean you should accept the null. As has been noted, current techniques are just unable to detect subtle cheating by GMs, and Niemann wo
932.
by gwern 3y ago
This version shouldn't be linked, even though it's the canonical one everyone links, because it's bad. If you check it against the original paper ( https://gwern.net/doc/cs/algorithm/1982-perlis.
933.
by gwern 3y ago
Amazingly enough, that pretty much works, but you'll never hear about it because the name of it is impossible to search for: https://www.wpeebles.com/Gpt
934.
by gwern 3y ago
Amusingly, _Catch-22_ is the only book to make both this list (which focuses more on prestigious or famous books rather than books anyone tries to read) and also my list of most-dropped books compiled from GoodReads statistics: https:&#x
935.
by gwern 3y ago
It is not 'balderdash', it is a reference to Scott's longstanding interest in Karl Friston & predictive processing paradigms in neuroscience & psychiatry. If you don't get the allusions, that's your problem,
936.
by gwern 3y ago
It is still pretty bad. When you look up old famous crimes, they're usually in there basically for life. Leaving aside cases like the 'I don't like Mondays' girl (44 years and still in there; temporal lobe lesion, possib
937.
by gwern 3y ago
The conversion is still very error-prone. It can't convert a lot of packages, and the last paper I read, StarVector, half the HTML version is just missing. (I think it hit an error at a figure of some sort.) I reported an error, but I&
938.
by gwern 3y ago
The standard AlphaZero doesn't handle that. For that you'd need to graduate to more complex variants like the aforementioned ReBeL, AlphaZe* https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10213697/ or
939.
by gwern 3y ago
Leon Bottou isn't a humanities professor, but a ML researcher. In fact, not just any ML researcher but arguably one of the ML researchers who most anticipated the current DL scaling era. Bottou was arguing for the virtues of SGD on t
940.
by gwern 3y ago
Related: https://news.ycombinator.com/item?id=38680239
941.
by gwern 3y ago
That's not a solution. You have to guard against all image URLs, because every domain and path can steganographically encode bits of information. 'foo.com/image/1.jpg' vs 'fo.com/img/2.jpg' just
942.
by gwern 3y ago
I think it's possible. One problem that I have with Remembrancer, as well as all approaches based on the same basic idea of 'show the user a list of similar items', is that this is fundamentally guaranteed to be mostly useles
943.
by gwern 3y ago
I wonder if even Tolkien's own works would preserve the 'essence' that people see in LotR... Imagine you took someone who had just read _The Hobbit_, and you gave them a copy of the _Silmarillion_ carefully scrubbed of any id
944.
by gwern 3y ago
I think the real rule for cats is, 'if your mother brought you it, then you hunt and eat it; if you feel sick in the hours after eating it, you stop eating it'. If you think about it, cats are such generalist obligate-carnivore pr
945.
by gwern 3y ago
> There’s a recent good book review summarizing this: https://www.newyorker.com/magazine/2023/11/20/charlie-chapli ... It's also a good example of how not to defend someone like Charlie Chaplin.
946.
by gwern 3y ago
You could also just steganographically encode it. You have the entire URL after the domain name to encode leaked data into. LLMs can do things like base-64 encoding no sweat. Encode some into the 'ID' in the path, some into the ca
947.
by gwern 3y ago
> Employees involved are well aware of the implications; I’ve seen conversations on Lark, ByteDance’s internal communication platform for employees, about how to “whitewash” the evidence through “data desensitization.” The misuse is so r
948.
by gwern 3y ago
No, it is not, and we are quite sure of this. The fatherhood correlation never survives any control for family-level factors, like comparing siblings or looking at kids who lose their father to accidents. And none of the evidence cited by O
949.
by gwern 3y ago
"In the next stage of the experiment, 85% of children and 95% of adults were also able to innovate...Effective tools were selected...75% by the best-performing model." "AI lacks the crucial human ability of innovation, resear
950.
by gwern 3y ago
I'd say, trying to read this, the biggest problems are: - tons of visual clutter, all those gradients and lines like the header or hero image - a floating ToC which insists on jamming in 'recommended links' (?!) the entire ti
951.
by gwern 3y ago
Before reading this essay, it may be useful to get some biographical background on the author, as she talks a lot about isolated snippets but otherwise provides no context for her life and how she got where she did: https://en.wi
952.
by gwern 3y ago
I can say that in all the reading I've done about Taoism, I've never seen this accusation. I bet OP can't name 3 examples of blaming wu-wei, by name, for 'Chinese laziness', which span at least 3 decades (much less
953.
by gwern 3y ago
I wonder if you could use AFL to try to fuzz your way to a shorter executable? There may be undocumented glitches or assumptions in how exactly the loader works which would shave some bytes off (and given how fundamental this is, any hits m
954.
by gwern 3y ago
I strongly disagree. If the report was guaranteed to be a whitewash (it may well turn out to be one, of course, but that's exactly what the current fight is over), the ex-Board & Shear & Taylor wouldn't be appealing to it
955.
by gwern 3y ago
If it's a gratitude tour for restoring Sam (or more cynically, probing everyone for their loyalties & attitudes in person), why isn't Sam the one doing all the meetings and in all the photos?
956.
by gwern 3y ago
I believe it. Note that this story is being sourced from Altman/Conway, even to the level of their private text messages. So they would have to be the ones fabricating this claim, but this story is embarrassing to them: if they were go
957.
by gwern 3y ago
One imagines they're a bit busy strategizing how to handle the regulators: https://www.bloomberg.com/news/articles/2023-12-08/ftc-makes... https://www.reuters.com/world/uk/uk-an
958.
by gwern 3y ago
As usual, a useless title. This reports a lot of interesting things, but who's going to read it with such a generic title? Overall: This one leaks heavily from the Altman/Conway camp but also from the director side, especially wha
959.
by gwern 3y ago
Ah, another entry for my list: https://gwern.net/review/cat#fuzz-testing
960.
by gwern 3y ago
That is disappointing, but I'm also a little curious - why did this go all the way to field trials if the gene insertion was completely screwed up, to the point of being on the wrong chromosome ? In November 2023, through molecu

More ›