Hacker Newsnew | past | comments | ask | show | jobs | submit | Thomashuet's commentslogin

Summary: We've use the most complexest, buzzwordiest training infrastructure to increase the performance of our base model by a whopping 0.5% (±1%).


But this isn’t about the performance, the infrastructure is the product here.


Indeed, most reliable way to make money in a gold rush is to sell shovels.


But is there such a thing as "modern standard Arabic"?



There is but nobody speaks it.


If it just means "don't take my guns", that's what they would write on their merch instead of some greek letters most people can't read. By using this Spartan phrase they're trying to evoke the mythical greatness of Sparta, and that is what this article is about.


"Come and take them" has long been a refrain of the gun crowd. This greek is a new version of the same thing.


Yeah, and everybody in Alabama always says "Go Team, Win Win Win" instead of "Roll Tide" because nobody can understand what "Roll Tide" means.

Nobody is actually invoking anything beyond the absolutely most superficial "Spartanness" and going deep into "Well, actually, the Spartans were not nice people if you dig deeply into the truth" is not relevant when nobody else is digging that deeply. To a first approximation nobody knows anything about the Spartans beyond "vaguely Grecian" and "good warriors I guess".

Despite all the sensitivity in the university system in the past 5-10 years, I've never even heard a whiff of a suggestion that Michigan State University needs to change its mascot... because nobody, not even the educated and politically sensitive, thinks anything about Spartans beyond "vaguely Grecian" and "good warriors I guess". If anyone had any sort of real knowledge about the situation it would have been considered extremely extremely insensitive... but nobody thinks about these things. Hard to get much more proof than that.


It sounds like this kind of essay isn't for you!

> "nobody knows anything about the Spartans beyond.."

"Sparta" is a "symbol" in our culture's "language of symbols" which stands roughly for "brutal and effective masculine warrior ethic, worthy of respect and fear". It is probably the single most unambiguous symbol available for this kind of thing, and is therefore used constantly—when somebody wants to make an argument like "might makes right", or "what if we didn't care about other people", or "men must be strong, austere, and violent to be worthy of respect", a reference to Sparta is never far away. Even when they don't make explicit reference to Sparta, it lends credibility to these sorts of ethics. It acquires its power from the fact that it actually existed in history and was seemingly successful, that it lives on with a generally admiring connotation, that it stands for the pinnacle of achievement in a certain sense.

The point of an attack on the historical accuracy of the symbolic Sparta is to weaken the power of the symbol to be used for evil. It did not exist as remembered, nor for all that long, it was not particularly effective, the admiration of Sparta has been concocted after-the-fact by commentators trying to fit it into their agendas, it was not as brutal as it is imagined to be, nor was it particularly important in history. It was just a weird and particularly unpleasant culture that existed at one point.

These are attacks on the structural integrity of a whole set of ideas which are regularly used to justify cruelty, abuse, and violence, and which mislead people—young men especially—as to the nature of virtue.

That it is sometimes used harmlessly doesn't have much to do with this at all.


Well written!


Thank you!


The main problem with this is that we're looking at the evolution of inflation-adjusted prices for common ressources. By definition, inflation-adjusted prices for common ressources must be constant, that's how we measure inflation. So it shouldn't come as a surprise that the prices do not change much in the long term.


That's obviously wrong. If one resource gets scarce but retains its use, it will get more expensive compared to other resources that remain plentiful.

All commodities together feed into inflation. So either some will become less expensive or some will outpace inflation.

What we're seeing here is that resource extraction can get better, so resources aren't really scarce. But there are certainly some commodities, not reflected here, that became much cheaper. E.g., corn or meat, so these metals might have become more expensive in relative terms.


I think you're right that the underlying problem is the economic theory and behaviour of nonrenewable resources. Or more accurately, the utter lack of correspondence between economic theory (Hotelling's Rule) and actual price behaviour.

This isn't a matter of inflation but of the lack of accounting for the fact that nonrenewable resource are, well, nonrenewable. Unlike (at a first blush) farming or labour, where utilisation currently doesn't mean that the same resource is unavailable in the future, once a nonrenewable resource is extracted and used, that production process cannot be run again.

Metals and ores can of course be recycled or reused, though in practice achieved rates are quite low (50% is exceptional, and that means that 1/32nd of the original resource is available after five cycles). Fuels which are combusted are a whole 'nother matter, as once burnt (or fissioned, in the case of nuclear power), the resource has been degraded and won't reform in anything less than geological time, if ever.

Once one comes to face with the fact that we're utilising fossil fuels at ~1--10 million times their rate of formation, the problem and inadequacy of economic pricing models becomes glaringly apparent.

Metal ores may recycle in less time, and some (e.g., iron) are massively abundant. But, again in the case of iron, when one considers that commercially viable ore deposits were formed 1--3 billion years ago, during the first phases of life on Earth, and driven by those biological processes.


You don't have to, you can run your favorite flavor of Linux. Unlike with the Mac Mini which can only run macOS.


Apple has never put any technical or legal obstacles in the way of installing other operating systems on Mac hardware. Nor do they assist in any way, it's consistent benign neglect.

The old Intel machines made excellent Linux boxes, excepting the TouchBar era because the TouchBar sucked (it was possible to install Linux, it would display the fake function keys, they worked, but not a good experience). I've converted two non-TouchBar Mac laptops into Linux machines, with zero complaints, one of them is in current use (not the laptop I'm typing on this instant however).

Now there's Asahi, which as a sibling comment points out, will surely be supported for M4 eventually. This is a great time to buy the M2 Minis and put Linux on them, if that's what you're into. Or you can wait around for the M4 port, whatever suits your needs.


Yet they made BootCamp. Do you see how foolish you look trying to defend nonsense?

Apple try to avoid being too heavy handed in the lockdown because they know the outrage it would cause from their legacy customers. Boil the frog slowly.

But they most definitely are trying to make the Mac more like an iPhone and they would rather not you install any other OS on it.

The bootloader not being completely locked is more for legacy reasons and multi-macOS support (dev/debug) but if you have any problem with it, you will need (surprise-surprise), another Mac for a DFU restore, just like an iPhone.


Asahi Linux has changed that equation for M1/M2 mac minis. I'm sure M4 will be supported soon.


Short version: a service known for evading YouTube's bot protection is complaining that ByteDance is bypassing their own protections. I agree that it's not nice from ByteDance but I find it hypocrite from Cobalt to call it evil.


> cobalt was created for public benefit, to protect people from ads and malware pushed by its alternatives

can't say the same for bytedance, which is designed to exploit users with various ads


It was created for donation money, lets not do mental gymnastics to justify one type of scraping and vilify another. Scraping is scraping and it's either all fair game or it's not all fair game.


I feel like you’re missing the point on purpose? Cobalt is asserting that it’s doing good based on the shadier behaviour of its competitors. But can you justify Cobalt in isolation any more than you can justify whoever was scraping it?


Yes.


You can't compare that... cobalt doesn't DDOS YouTube


Cobalt is also completely free, without ads or any other monetization besides donations, it's purely meant to help normal people download videos for normal people purposes. It's not like they're a for-profit data harvesting outfit complaining about getting abused by another for-profit data harvesting outfit.


You're just saying that Cobalt is small and non-profit so they must be good and YouTube and ByteDance are big and rich so they must be evil. But if you only look that what they are actually doing here, it's very similar: bypassing protections to use a service in a way that the service provider doesn't like.


Bytedance and youtube are evil, but not beacause they are big and rich. Cobalt is good, but not because they are small and a non-profit.


If bytedance are so big and rich why don’t they implement their own scraping solution instead of abusing a small service like cobalt.


...Because someone scraping from a Bytedance IP range is not necessarily Bytedance, just like requests from an AWS IP do not imply Amazon authored the spider


In isolation, a thief masquerading as a security system technician and an actual technician both do good work by checking on your home security. You can't meaningfully say one is better than the other, because even though one is secretly casing out your home so he can rob it later, in isolation they're doing the same thing.


Cobalt is bypassing protections to allow legitimate Youtube users to download single videos without causing harm and with no monetary incentives. Bytedance is mass downloading thounsands of videos, all for monetary incentives while heavily breaking the TOS and potentially ignoring copyright laws. Similar, but one is doing way more harm than the other.


> and with no monetary incentives

Donations are a monetary incentive

> while heavily breaking the TOS and potentially ignoring copyright laws

Cobalt also breaks the TOS and ignores copyright laws, personally I don't think that matters but having a double standard when one company does it "It's ok when they do it" and when one you don't like does it you try to use copyright laws and TOS as a weapon just makes me think it really isn't about TOS or copyright is it.

Also just gives YouTube ammunition to impose stricter protection against smaller violators like cobalt, like self running yt-dlp


Cobalt didn’t say the DDOS was evil, they said:

“bytedance's scraper was specifically built to go around cloudflare & other web security solutions, which is just genuinely evil”

So I would say it’s a fair comparison.


> built to go around cloudflare

Then they either didn't set up CF correctly or they just use the mode in most headless browsers that bypasses default CF protection when CF is not in attack mode.


I don't see the hypocrisy here. Cobalt is a small, free service that results in Google (or so the argument goes) making less profit. ByteDance are a giant money printing machine using that free service for their own ends. They have more than enough resources to not abuse a free one.


Let's say hypothetically Cobalt was made by ByteDance as a way to scrape youtube and have a scapegoat. Is it still okay?

If your opinion changes because the owner is different, even though the service stays the same, that's hypocritical.


Of course it isn't hypocritical. It's like the old story of the poor man stealing bread to feed his starving family, of course circumstances matter. It's silly to suggest otherwise.


Using Cobalt doesn't do anything to feed their family. It's an old poor man stealing a blu-ray DVD vs a rich young man stealing a blu-ray DVD.


Apples to oranges - abusing an undocumented API of a foreign service to mass-scrape another one by proxy is not the same as sending singular, user-created requests.


Their claim that nothing tells you the email corresponds to the new issue is wrong, the "(Issue #1)" in the title means exactly that. I have actually received the same email myself and immediately recognized it as a new issue created on the repo. This user is obviously not used to GitHub issues as is made clear by the fact that this is the first issue on this repo. I guess GitHub needs to do a better job teaching new users.


True, but I have worked at companies who employ users that maybe aren't entirely up to speed on the technical details and they have GitHub account's for submitting bug reports. This would very easily fool some of these people.

Technical people might spot this, but that also isn't a free pass for GitHub to not do better here.


I don't understand the use of a textual description. In which scenario do you not have enough space for a lens and yet have a textual description of the scene?


> Tonal languages allows individuals to express way more than Latin based languages.

Do you have any evidence of this? I've never heard this claim before.


I don't see how opt-in solves the problem of people who content they don't own. They could opt-in your content just as they can currently post it on their website.


At least, with opt-in, you would be able to prove that you were allowed to train your model on the content. With opt-out, that is impossible: You can not prove that something does not exist.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: