Generally it's configurable. The one that comes to mind first is TokenWorks's Anti-Passback feature which says "Set your custom timeframe (1 hour to 7 days)"
But if they want to design for extreme probabilities you need to account for tail risks, so their design should have included a missile defense system.
At some point you need to start worrying about asteroid defense too.
That depends on the data. If this is EBS or single-AZ S3, then from Amazon's perspective this was correct. Backup responsibly (for any data that does need to be backed up) lives with the customer, and Amazon has no way of knowing about that. EBS data data is unrecoverable, and that's what's reported.
Now if this was multi-AZ S3 or whatever then this would be significant.
The article does not tell us what products were impacted.
I was unaware that Amazon even sold single AZ S3. 20% discount. Doesn't seem worth it. By the time I commit to purchasing S3 space, it has to be important data.
I get that S3 is convenient and reasonably performant, but it is not cheap at all.
Google internally has lots of possible redundancy levels for data.
They don't sell any of the lower and less reliable levels to the public, I suspect simply because the reputational damage from losing user data is so bad, and the news will take no notice of the fact the user got a discount for less reliable storage.
Most of Google's customers wouldn't know how to choose anyways, if these were exposed. My memory was quite hazy but I recall having a discussion with my colleague on choosing which Reed–Solomon code for our Colossus files, and apparently the choice was down to RS(8,3) or RS(9,3). I don't think even as Googlers we really had enough information to make an informed choice. Comparatively it was much easier to decide which cells to use for multi-location replication in Placer.
It's a great service for large caches. For example, we process a lot of imagery that we download from third-party providers. We save a lot of latency by storing the data in a single-AZ S3.
If it dies, we will just have to re-download the data.
> EBS data data (sic) is unrecoverable, and that's what's reported.
I don't see where this is reported? TFA does not mention EBS. In fact, TFA seems to be nigh content-free, beyond "AWS (allegedly, and is uncited) says they cannot restore some data."
> The article does not tell us what products were impacted.
… right … which conflicts with EBS being what's reported …
(I would agree with your point that if EBS, or some AZ-level data was lost, then, yeah, that's the contract.)
This is the inevitable result of having hundreds of thousands of cameras all collecting data and searchable from one point. Even if every search required a warrant, this would still be abused with cops lying on affidavits etc.
We don't need this system, we don't need to tolerate having this vector for abuse lying around.
No. The recent decades has shown that there are no laws for a sufficiently motivated entity or government. They're just words on paper, a promise. You can't expect them to keep that promise indefinitely, especially when those incentives you speak of (profit for the companies, easy repressions and surveillance for the government) point in the opposite direction. The only way to ensure this doesn't happen is to not have the tools of repression to be available at a second's notice to them. We live in a world of power, not of laws, at least as far as the most powerful are concerned.
Besides, I don't want them to have a right to track and correlate data about my location even if it's the best government ever that only ever uses this to solve the crimes that I agree are bad. No thanks.
The precedent on this matter is surprisingly thin on the ground, but the legal system has declined to agree with "their store, their rules" for web sites, with a particularly interesting precedent on that matter from "Sandvig v. Barr, 451 F. Supp. 3d 1 (D.D.C. 2020)" where the District of Columbia in 2020 said "Criminalizing terms-of-service violations risks turning each website into its own criminal jurisdiction and each webmaster into his own legislature. Such an arrangement, wherein each website's terms of service “is a law unto itself,” would raise serious problems. This concern, then, supports a narrow interpretation of the CFAA." https://opencasebook.org/casebooks/1797-tlc-casebook/resourc... (paragraph 66)
But all we have is that and a couple of rulings from the Ninth Circuit, one of which is the HN link. For something this critical I'm surprised. I remember writing about this in 2001 or so and you'd think for something so consequential we'd have more lawsuits to point at by now. One wonders how many companies have considered it, looked at the thin precedents we have, and realized that while the victory would be slightly advantageous writing a loss into precedent would be completely devestating and declined to roll the dice.
Maybe a better analogy: They can ban you from going through their drive through on a bicycle, but can they sue bicycle manufacturers because bike riders are going through their drive through?
In this analogy, it seems the originating court said yes, they can sue the bicycle manufacturer (in the analogy), but on appeal the answer was no, they cannot. That seems like the appeals court got it right to me.
I'm not familiar with the case, but to me it sounds like the bicycle manufacturer is encouraging people to use bikes in the drive through, maybe even providing specific tools explicitly for that purpose?
> If I'm busy, I can hire a person to go stand in line and buy the device (me asking Comet to shop on my behalf)
I feel like this is already ceding too much ground. A better analogy is "can I hire someone to read this product catalog I got in the mail for me." My browser isn't Amazon's store, it's a device for reading catalogs they send me.
3. You can walk into an Apple store and buy a phone without looking at their PCs, tablets, wall of accessories, etc. Apple cannot force you to look at everything else in their store.
Similarly, Amazon cannot force you to render the content they deliver to you. They can absolutely control access to their content, but they cannot control if you actually see the content they are delivering to you.
Amazon is a party to interactions with their website. That gives them standing to sue over those interactions. Whether they can sue Perplexity specifically is what was being debated in these motions.
Nah, once you have the bits on your device you should be free to do with them whatever you want. Amazon doesn't get to dictate how you consume what they give out for free.
What you think you should be able to do with the data you get back from contacting Amazon’s servers doesn’t impact whether Amazon has standing to sue you over how and why you contacted their servers.
Sure and what you comment doesn't change that people have opinions about the validity of such cases. Amazon cares more about their bottom line than the health of their workers, we should be highly skeptical of anything they want to do as there is a strong chance that it will hurt + immiserate workers.
Not all of us have the same bathroom fetish as Bezos.
Sure it does, you're making a lot of assumptions here. The most egregious one is believing that that I think the legal system means anything to actual justice and not a system that is mostly used to protect the interests of the elite at the expense of the poor.
In cases like these, companies often try to go after the upstream project, not just operators. The law gives them lots of tools for this: the CFAA has a conspiracy provision; The DMCA has a provision against making software for circumventing copyright; Trademark law might make the name nitter or xcancel illegal; and finally even if the claims are weak the litigation itself is an enormous burden
I noticed Google AI Mode (so Gemini, I was doing some quick research in the browser ok) got a detail wrong once, so I asked it what happened. I kept digging deeper and finally just asked it to write me a Python script visualizing what happened. It did, complete with vectors.
Now I want to go find that conversation in my history and see if it can tell me about weights, and how that contributed.
Distillation does not reveal the weights, it produces a different network with similar behaviour. Weight space isn't even identifiable: permutation and scaling symmetries mean many weight sets give the same function.
The model also lacks the machinery. No training loop, no gradient descent, nothing to write to.
And a model only sees its own sampled tokens, not the distribution behind them, which are possibly filtered or post-processed. Distillation from that works but is less sample-efficient than soft-label distillation.
They usually don't, but if they break out and take over the network of the company, it becomes possible to reach around and grab them. This kind of break out has happened, though I don't know of any weights being nabbed.
This is one area where I thinks self-report is probably fairly accurate. The study also cites some metastudies of other research that shows the same correlation (I don't have access to see if those are based on self-reports)
Especially since I think there's a pretty easy to understand mechanism for a hot drink to cause cancer.
Hot liquid damages the cells lining your esophagus which potentially damages the DNA of some cell leading to cancer. Pretty much exactly what hard liquors do only using heat instead of ethanol to damage the cells.
reply