yorwba

40 ms·

961.
by yorwba 1y ago
Their architecture uses a mix of Transformer and Mamba layers. The question isn't whether it will replace Transformers, but whether it'll become part of the toolkit or whether it'll get abandoned like many other promising a
962.
by yorwba 1y ago
Currently, less than 70% of the world population use the internet. Universal adaption may be inevitable, but it could take a few more decades. Less than 40% use Facebook at least once a month. Comparable user numbers for LLMs are a bit hard
963.
The gains from trade are not the gains from trade(nicholasdecker.substack.com)
5 points by yorwba 1y ago|0 comments
964.
by yorwba 1y ago
It's possible to identify a surprisingly large number of matching words by learning a linear transformation mapping word vectors from two different languages into the same space (e.g. https://arxiv.org/abs/1805.062
965.
by yorwba 1y ago
The author seems a bit too excited about the discovery that the dot product of the vectors [ a , b ] and [1, 1] is a + b . I don't think the problem with getting neural nets to do arithmetic is that they literally can't add tw
966.
by yorwba 1y ago
Looking it up, it seems like the CEO actually got raided by the FBI last year: https://www.reuters.com/world/us/fbi-raids-polymarket-ceos-h... So maybe the wheels of justice are just grinding a bit slowly. The ter
967.
by yorwba 1y ago
The spigot doesn't seem to distinguish between crawlers that make more than 15 requests per second and those that make less. I think it would be nicer to throw up a "429 Too Many Requests" page when you think the load is too
968.
by yorwba 1y ago
People use crypto on Polymarket because it doesn't comply with gambling regulations, so in theory isn't allowed to have US customers. Using crypto as an intermediary lets Polymarket pretend not to know where the money is coming fr
969.
by yorwba 1y ago
Yes, it's difficult to detect whether something is enemy propaganda if you only look at the content. During WWII, sometimes propagandists would take an official statement (e.g. the government claiming that food production was sufficien
970.
by yorwba 1y ago
No, Wu-Tang ultimately derives from the Wudang Mountains, with the corresponding Cantonese being Moudong https://en.wiktionary.org/wiki/%E6%AD%A6%E7%95%B6%E5%B1%B1
971.
by yorwba 1y ago
If you click on the time next to the username, you get a dedicated page for that comment with additional options: | parent | context | flag | vouch | favorite | ... but you need some minimum karma for the "flag" and "vouch&qu
972.
by yorwba 1y ago
Which means that whoever is responsible for updating https://github.com/xai-org/grok-prompts neglected to include Grok 4.
973.
by yorwba 1y ago
The user did not ask for Musk's opinion. But the model issued that search query (yes, using the standard Twitter search syntax) to inform its response anyway.
974.
by yorwba 1y ago
The attack does not require a fixed point of the hash function to be integrated into the program, it merely involves an implementation of the hash function included in the program, being fed the exact same input as the hash function used
975.
by yorwba 1y ago
Figure 21 shows that initial implementation time (which I take to be time to PR) increased as well, although post-review time increased even more (but doesn't seem to have a significant impact on the total). But Figure 18 shows that ti
976.
by yorwba 1y ago
The time is given as "last October," and the work they were unfamiliar with is presumably "contrived proof protocols that are vulnerable to attack, no matter which hash function you use" as stated in the immediately prec
977.
by yorwba 1y ago
If the target is parity with CAM16-UCS, CAM16-UCS is best, tautologically. Sure, if you need a fast approximation, by all means fall back to Oklab, but that optimization isn't going to be necessary in all cases.
978.
by yorwba 1y ago
The nuanced take was also discussed here at the time: https://news.ycombinator.com/item?id=42939312
979.
by yorwba 1y ago
Oklab is not perceptually uniform. It's better than other color spaces with equally simple conversion functions, but in the end, it was created as a simple approximation to more complex color spaces, so compared to the best you could d
980.
by yorwba 1y ago
That density is the volume density. For most applications, you care about the density in terms of number of hyperspheres instead. The largest hypersphere that fits inside an n-dimensional unit hypercube has radius 1/2 and volume π^(n&#
981.
by yorwba 1y ago
It does test for AGI. For its absence.
982.
by yorwba 1y ago
There are no errors in the calculation, but it's wrong anyways because it calculates the answer to the wrong question. "At best" suggests this is the largest possible effect, but it is the smallest possible. To get an upper b
983.
by yorwba 1y ago
I think the people behind the ARC Prize agree that getting a high score doesn't mean we have AGI. (They already updated the benchmark once to make it harder.) But an AGI should get a similarly high score as humans do. So current models
984.
by yorwba 1y ago
It's true that the last layer's output for a given input token only affects the corresponding output token and is discarded afterwards. But the penultimate layer's output affects the computation of the last layer for all futu
985.
by yorwba 1y ago
In this case, the lookup table is used for popcount, and there's a comment in the Z80 assembly that says "One lookup vs eight bit tests." If the code could make use of a hardware popcount instruction, the lookup table would l
986.
by yorwba 1y ago
Usually, everyone does do that, which is why only hard-to-detect patterns remain profitable. Not something obvious like "buy options in the morning and sell in the evening" as in this example. But maybe Jane Street only traded l
987.
by yorwba 1y ago
I think the reporter's motivations would've come across more clearly if you had posted a paragraph-by-paragraph translation instead of the current abridged version. (I assume Dilemma Works is your Substack.) Lots of details that a
988.
by yorwba 1y ago
I think the case you're talking about is this one: https://arstechnica.com/tech-policy/2024/10/bytedance-intern... where it was a ByteDance intern.
989.
by yorwba 1y ago
The original whisteblower article in Chinese at the bottom (but not the English version at the top) has this part: 实际上，对于后续训了很久很久的这个模型，Honestagi能够分析出这个量级的相似性我已经很诧异了，因为这个模型为了续训洗参数，所付出的算力甚至早就足够从头训一个同档位的模型了。听同事说他们为了洗掉千问的水印，采取了不少办法，甚至包括故意训了脏数据。
990.
by yorwba 1y ago
I think he just made a self-aware observation: noticing a trait being unusually common among criminals he investigates makes him subconsciously associate it with crime even in the general population. Then somebody decided to translate &quot

More ›