yorwba

36 ms·

901.
by yorwba 1y ago
If you completely do away with autoregression, prompt tokens can pay attention to generated tokens, so even the prompt tokens' KV vectors change at every step and you cannot cache anything. For this reason, models that generate text us
902.
by yorwba 1y ago
> It's hard to tell where the prompt stops and the CoT response starts, in fact. That's because you're looking at the final output that includes neither the prompt nor the intermediate chain of thought.
903.
by yorwba 1y ago
In a sense? As the article notes, models trained using standard attention develop attention sinks naturally and removing them makes the model deteriorate completely, so the hallucinations you're thinking of were most likely output by a
904.
by yorwba 1y ago
The point is that there's not always a right token to attend to. If the information you're looking for is not there, no clever attention scheme will find it. The best you can hope for when that happens is that the value returned i
905.
by yorwba 1y ago
In April, China issued window guidance to restrict datacenter construction to avoid overbuilding: https://www.cls.cn/detail/2005527 The article you link mentions similar capacity cuts in the car industry, but apparentl
906.
by yorwba 1y ago
During my undergrad in Germany, the CS department was in the process of switching from optional homework to various forms of mandatory homework (either directly counting towards the final grade, or requiring a minimum score on the homework
907.
by yorwba 1y ago
You could have a company-provided Anki account for each user where you add and remove cards just for that user. (I thought you might even be able to use your own server, but that doesn't seem to be an option for the iOS app: https:&#x
908.
by yorwba 1y ago
It's on page 14 of the technical report. They generate synthetic data by putting text on top of an image, apparently without taking the original lighting into account. So that's the look the model reproduces. Garbage in, garbage o
909.
by yorwba 1y ago
If the application is simulating a crisp higher-resolution image that was slightly blurred while downscaling to the output resolution, a linear color space is exactly the right choice. Yes, it means bright objects on a dark background will
910.
by yorwba 1y ago
See this other submission by the same user: https://news.ycombinator.com/item?id=44787837
911.
by yorwba 1y ago
All three claims of gold medal performance on IMO 2025 that I'm aware of solved the first 5 problems, that were designed to be solvable by application of standard techniques, but got stumped on the sixth problem that was a bit more unu
912.
by yorwba 1y ago
The article isn't using the graphs to talk about a difference between 2015 and 2017, but about the overall level, so it would indeed be surprising if there had been a large change between those years. But there are several demographic
913.
by yorwba 1y ago
The source is linked under each graph in the article.
914.
by yorwba 1y ago
You might be able to build something similar yourself using declension data extracted from Wiktionary using wiktextract: https://github.com/tatuylonen/wiktextract#pre-extracted-data
915.
by yorwba 1y ago
> [2] In the comments, zero_k posted a link to the SAT competition's parallel track. The 2025 results page is here: https://satcompetition.github.io/2025/results.html Parallel solvers consistently score lower (
916.
by yorwba 1y ago
When it comes to prioritizing capital deployment for education, it's probably better to start with primary schools before thinking about top universities. https://wid.world/news-article/china-vs-india-how-human-cap
917.
by yorwba 1y ago
It sounds like the books were collected as evidence during a search. All kinds of things can serve as evidence, even if they're not themselves illegal.
918.
by yorwba 1y ago
> In theory today, I should be able to process <long quote from document> <specific query> and just stop after the long document and save the KV cache right? People do this, it's called prefix caching. There's also
919.
by yorwba 1y ago
The "Durability Against Impact and Fatigue" section is fun: "[...] mechanical damage rendered it unmountable, but the tactile sensing still worked and could be used to control the UR5. The device ultimately failed when additi
920.
by yorwba 1y ago
> Searched the publishing institute (KAIST) website with google search You could've skipped this part by scrolling down and following the "More information" link. > Is this enough fact-checking Depends on which fact you
921.
by yorwba 1y ago
I think the data set is generally considered to consist of the images, not the list of links for downloading the images. That the data set aggregator doesn't directly host the images themselves matters when you want to issue a takedown
922.
by yorwba 1y ago
It seems like the gimmick of Ensō is that it's write-only and doesn't allow you to edit anything, so not being able to see your typos might be for the best actually.
923.
by yorwba 1y ago
> The German constitution guarantees Freedom of Speech? Article 5 of the Basic Law guarantees freedom of expression, freedom of the press and freedom of reporting by broadcast and film. It immediately restricts those freedoms with "
924.
by yorwba 1y ago
> Governments now want to pretend that it matters where the user is rather than where the site is This is not a new thing either. Whenever something somehow touches multiple jurisdictions, it's generally safe to assume that laws fro
925.
by yorwba 1y ago
There's not a trend against freedom of speech so much as existing laws outlawing certain categories of speech being applied to the internet. If you lie in a commercial context, that's fraud; if you lie in court, that's perjur
926.
by yorwba 1y ago
It is totally the advice I want to be giving. Given the choice between an echo chamber matched to my interests and wading through a stream of unfiltered crap, I'll take the echo chamber every time. (Of course there's also the opti
927.
by yorwba 1y ago
Peer review is a way to distribute the work of identifying which papers are potentially worth reading. If you're starting from an individual paper and then ask yourself whether it was peer reviewed or not, you're doing it wrong. I
928.
by yorwba 1y ago
This HRM is essentially purpose-designed for solving puzzles with a small number of rules interacting in complex ways. Because the number of rules is small, a small model can learn them. Because the model is small, it can be run many times
929.
by yorwba 1y ago
That makes it popular for a movie . It also means that most people didn't watch it, at least not in theaters. (I guess that would be true even for the most popular movie of all time. The most popular thing in some category is rarely
930.
by yorwba 1y ago
The part that is in principle amenable to replication is where they throw a lot of stuff at the wall and see what sticks. The part where they hype their own work, on the other hand... as a rule of thumb, if this really were a breakthrough o

More ›