yorwba
yorwba on Hacker News
You're probably thinking about pre-election polls. Those are frequently off by a few percentage points even if they took great care to take a large representative sample of the population because it's difficult to predict turnout. However, there's one opportunity to get a representative sample of voters: by conducting an exit poll right after they voted. If exit polls show a large discrepancy from actual results, it's usually seen as an indicator of fraud.
I don't see where school libraries are mentioned.
The author bio at the bottom says "Hazel (the artist) and Bell (the writer) are an English-speaking duo behind the webcomics "Electric Bones" and "Always Raining Here"." Which are up
https://electricbonescomic.com https://www.alwaysraininghere.com I think they just decided to blow their self-promotion opportunity for an inside joke.
The dedicated fanbase, sure. But not people who don't actively seek out sexy raccoons, although they might read the comic if it shows up in their feed when they're bored. So not being able to post on popular platforms costs them a customer-acquisition funnel.
The usual solution is to put conspicuous censorship markers over everything the platform finds objectionable and then let the audience know where they can buy the uncensored edition.
Not replacing expensive infrastructure before it really needs to be replaced is simply the economically better choice. If you replace it earlier, you have all the same issues with not being able to use it while construction is ongoing and the same costs and so on, so all you achieve with a replacement ahead of schedule is to artificially cut short the lifespan of perfectly usable infrastructure.
document.querySelector("table").style["table-layout"] = "auto";
> if any pair of countries have conflicting rules then you simply can't expose your service to both
Yes. This is also how physical products work. The internet makes it easy and the default to offer a service worldwide, hence also making it easy to get in conflict with the law worldwide. You have to pick and choose which laws to break, which laws to comply with by minimizing exposure, and which laws to comply with by making a tailor-made version for a specific market.
If there's no point in having the best model, there's also no point in leapfrogging the current best model. Conversely, labs trying to leapfrog each other indicates that they consider the advantages of having the best model (or at least a better one than they had before) worth it, even taking the competition into account.
You don't need to increase the number of vertices to get more balanced triangles. None of the methods in the article add vertices, but the Delaunay triangulation is better than the others at creating balanced triangles. It's just also the slowest, so there's a tradeoff between triangle quality and speed.
"Comic illustrated by Stephanie Cortes of Ibis Craft.
Poster and comic story by Michal Zalewski (@lcamtuf) of Calif."
Surely all the websites that had scheduled downtime for the patch are no longer vulnerable? Would be kind of pointless otherwise.
That doesn't stop you from asking for the certificate on a separate connection with your own session key, and terminating the original connection if you don't like the certificate.
According to that Reuters article $72B of the $80B loss is an "accounting charge reflecting an increase in the estimated value of bonds that could eventually convert into Anthropic shares," i.e. they borrowed money in return for stock options. That's bad for Anthropic investors, but essentially free for the company.
Their revenue "more than doubled from the first quarter to $11.6 billion in the second quarter" which is less than almost tripling so first quarter revenue might've been $11.6B/2.5 ≈ $4.6B for $16.2B total revenue in the first half. To get that revenue, they "spent nearly $9 billion on computing capacity and infrastructure in the first half." Scaling up for the $16.8B third-quarter revenue suggests $9.3B compute spend. Meanwhile to get the $8B remaining first-half loss, they must've had $24.2B expenses, $15.2B of which are not compute, so about $7.6B per quarter. If that stays fixed, total third-quarter expenses should be about $16.9B, fairly close to revenue. I think reporting that their second quarter was profitable focused on the operating income exceeding operating expenses only.
So there are actually several senses in which Anthropic could be profitable or fail to be:
1. Are operations bringing in more money than they cost? (Supposedly already happened.)
2. Are operations bringing in enough money to fund capital expenses? (Maybe close, maybe barely so.)
3. Does investing in Anthropic yield good returns?
Note that the last point crucially depends on their stock price. If it drops a little, those "bonds that could eventually convert into Anthropic shares" start showing up on the positive side of the balance sheet instead. Insofar as the stock market is efficient, if they IPO, it will be priced such that it's no longer obvious whether it's better or worse than any other investment. But early investors and founders will make bank.
From your link:
"Open-source technology and research already in the public domain would not be covered, and the government will provide streamlined procedures, such as automatic licensing, to exports involving open-source AI technologies."
China is making companies tell them ahead of time before they release large, capable models. It's not a ban yet.
If you look up something on Wikipedia and include it in your work
without citing it, it's an even worse pathway to citogenesis, since someone trying to check whether you can be used as a source for Wikipedia won't know that that's where you got it from in the first place.
Of course for most things you should have no need to refer to Wikipedia at all, and hence also no need to cite it.
They were a newsworthy company when their domain got shadowbanned, too... The spam detector is blind to such things.
How did you determine that it has been "flagged by many"?
They're still going through their logs from June. These are not new incidents, but newly-detected incidents. Whether or not this means they can now detect it while it's happening is unclear.
When you control the content, I agree that ruby tags would be better, since they can easily handle alternative readings and also enable smarter automation to infer the correct reading from context.
But you could set this as the default system font for Japanese and enjoy furigana everywhere, even in places that don't support ruby tags at all.
Declaring them a problem and then shooting them is the contingency plan.
Presumably they registered, but that doesn't mean the model passed through export review. It just means that some bureaucrat now has the opportunity to check the product characteristics they reported and decide whether that kind of thing should be allowed for export or not. (Though with the weights out in the open, that's maybe a bit pointless.)
Having installed WPS Office and read the thing (PDF uploaded here:
https://upload.wikimedia.org/wikipedia/commons/8/8b/%E6%8B%9... ) I can say that the draft definitely needs some work, since e.g. it prohibits deepfake detection technology with precision and recall above 95%, but merely restricts it if precision and recall are above 98%. Presumably it was intended the other way around.
Footnote 2 promises automatic permission for open source, hosting on companies' own data centers abroad, cross-border chip manufacturing, joint R&D with foreign resources, and international standardization. This probably refers to Article 14 of the Foreign Trade Law. https://en.moj.gov.cn/2023-12/15/c_948360.htm#:~:text=to%20m...
So companies can continue to release weights for models with more than 10^26 FLOPS, 2T parameters or that rank in the top 10 worldwide, but they need to report this beforehand to receive their automatic export license, which might then get revoked if the department of foreign trade feels like this kind of capability shouldn't be given away for free.
That section shows a single example, where pixels in some rectangular region around the selected object are being changed. For background removal, that rectangular region would encompass the entire image, so it's not clear whether you have any particular advantage there.
