mikehearn

"I need to let you know that we are unable to issue compensation for degraded service or technical errors that result in incorrect billing routing."

Not sure I've ever seen a company openly take this position. This is a crazy policy.

It's a fine line that's been drawn, but this ruling says that AI can't own a copyright itself, not that AI output is inherently ineligible for copyright protection or automatically public domain. A human can still own the output from an LLM.

The argument is that converting static text into an LLM is sufficiently transformative to qualify for fair use, while distilling one LLM's output to create another LLM is not. Whether you buy that or not is up to you, but I think that's the fundamental difference.

This is a well known blindspot for LLMs. It's the machine version of showing a human an optical illusion and then judging their intelligence when they fail to perceive the reality of the image (the gray box example at the top of https://en.wikipedia.org/wiki/Optical_illusion is a good example). The failure is a result of their/our fundamental architecture.

I still can't believe the guy went to Indonesia, went into the monkeys' habitat, gained their trust, set up the camera on a tripod in a way the monkeys would have access to it, adjusted the focus/exposure to capture a facial close-up -- basically engineered the entire situation specifically for that outcome, and simply because he didn't physically hit the shutter he lost credit for the photo. Meanwhile I can open my phone's camera, spin around three times, take a photo of whatever the hell happens to be in its viewfinder and somehow that is sufficient human creativity to deserve copyright protection.

Aggregate data from the Bureau of Labor Statistics.

https://www.in2013dollars.com/Food/price-inflation/2019-to-2...

https://www.nerdwallet.com/article/finance/price-of-food

It's closer to 28%. I wrote the initial post from my memory of the stat, which is why I approximated it.

To me, this is at the heart of why Trump won this election. I honestly do not believe your grocery bill has tripled. That's 200% inflation, which is an insane number. The statistics we have are that groceries have gone up ~25%. I have such a hard time imagining any combination of products that would add up to 8x the national inflation average of groceries.

But, I also don't think you're lying. I think you honestly believe your grocery bill tripled, and I think a lot of people have a similar internal impression about how bad inflation got. It's not useful for me (or, for politicians) to try and argue it logically. No one can check your receipts from 2019 and 2024 and say, look, things aren't actually that bad. Dems needed to kind of take it at face value and come up with a solution to something that people feel is real, and they just did not do that.

Editing to add: I might as well add the lowest effort source to the ~25% number, which comes from using the search feature of ChatGPT (sorry). https://chatgpt.com/share/672b7e09-4b58-800e-a3df-58f38c33bc...

As someone with the same name as a somewhat well-known former Bitcoin developer, this is sort of a latent fear I have. I would expect that someone dumb enough to think a home invasion is a good idea is also dumb enough to not double-check whether they've got the right guy.

It's extremely weird to see this site on HN! I built this site in 2014 -- and haven't touched it since. I wasn't a developer then, I was a product manager, and this was a "look, hiring managers, I can build things" side project (it worked, I've been a dev since 2016).

Despite being about 40% broken I keep the site up because it's still reasonably functional and there are a surprising amount of sites that now depend on having hotlinked the patterns directly from this domain. If it ever degrades to the point of being actively dangerous (and the attribution link rot is pretty close), I'll shut it down. Until then, it's a fun relic from the internet of a decade ago.

Just to answer a question upthread (and I 100% agree this should be on the website), the patterns are all CC-BY-3.0, meaning it just requires attribution and any pattern can be used for free.

"Every artist, performer and creator on Patreon is about to get screwed out of 30% of their gross revenue"

Does Apple have access to Patreon creators' gross revenue? I thought they only charged commissions on payments through IAP, which I assumed is only a minority of their overall gross.

I can be that guy. I use Rewind for Mac, which is almost identical to Recall in functionality. I love it, and I've used it frequently to find things that otherwise would have been lost forever.

Most recently I used it to refresh my memory on a particularly convoluted way to authenticate with a third-party oauth system (it involved using an online oauth debugger and curl commands). I had gone through the process once successfully weeks ago, but by the time I had to do it again I'd forgotten every detail. Rather than have to go through the process of figuring it out again, I went back to my successful attempt, watched it, and basically retraced my steps. Rewind probably saved me an hour or two.

My take on Recall is that, like with almost everything, it's a trade-off of security for convenience. I find it valuable enough that I'm willing to make the trade-off, but others might not.

I'm trying to square the claims in this article with what Microsoft says.

Article: "This database file has a record of everything you’ve ever viewed on your PC in plain text"

Microsoft: "Snapshots are encrypted by Device Encryption or BitLocker, which are enabled by default on Windows 11."

https://support.microsoft.com/en-us/windows/privacy-and-cont...

The article is a little bit hand-wavy about how exactly the database comes to be decrypted and remotely exfiltrated. The headline says it takes "two lines of code" but unless I'm missing it, I don't see those lines discussed in the article.

InstantID uses a non-commercial licensed model (from insightface) as part of its pipeline so I think that makes it a no-go for being part of Stability's commercial service.

> A criminal who is otherwise not legally able to acquire or possess a firearm is not going to care about 3D printer laws. This only impacts the law abiding.

This is kind of a blanket argument against all laws, right?

The original implementation also involved sending a "safety voucher" with each photo uploaded to iCloud, which contained a thumbnail of the photo as well as some other metadata.

The vouchers were encrypted, and could only be decrypted if there were, I believe, 30 independent matches against their CSAM hash table in the cloud. At that point the vouchers could be decrypted and reviewed by a human as a check against false-positives.

It sounds like with a raw byte hash they might be able to match a photo against a list of CSAM hashes, but they wouldn't be able to do the human review of the photo's contents because of E2E.

The article seems to make that claim:

"A separate version for the Apple Watch would remain [in the App Store], but then Apple pulled that one as well, telling Eleftheriou that keyboards aren’t allowed on the Apple Watch."

> A separate version for the Apple Watch would remain, but then Apple pulled that one as well, telling Eleftheriou that keyboards aren’t allowed on the Apple Watch.

This is wrong, as far as I can tell. The watch app is still in the App Store.

https://apps.apple.com/us/app/flicktype-watch-keyboard/id135...

I just installed it and verified that it works, although the Watch keyboard itself is hidden behind a $10 in-app purchase.

This analogy doesn't accurately represent the technology, at least as I understand it.

In Apple's implementation, the device never knows if a particular picture is a CSAM match. That determination is made in iCloud when the server attempts to decrypt the safety voucher. Until that point, it's just an encrypted payload that the device can't interpret one way or the other.

In your analogy, where "your home" is the equivalent of "your device", the police never enter the home to determine whether you have anything illegal. Instead, there's some process that boxes up all your stuff into nondescript, anonymous boxes that can only be opened if someone has the key.

To determine illegality, you'd have to voluntarily send them off to the police (police = iCloud), where they only have a handful of keys - they have a "gun" key, a "knife" key, and a few other keys for boxes containing illegal items. But the boxes are nondescript, so the police don't know whether you have anything illegal until they insert the key and turn it. If the "gun" key successfully opens the box, the box contains a gun, and you are reported. If all the police's keys fail on a particular box, then whatever is inside must not be illegal and the police never learn its contents.

Needless to say, this analogy is tortured because it's hard to apply Apple's tech to a physical process, but the point is that whether something is "illegal" isn't able to be determined until you voluntarily ship it off to an entity that has the keys to unlock it.

Isn't the "interpreting" step the one that matters?

Apple takes a photo, runs it through some on-device transformations to create an encrypted safety voucher, then it gets "interpreted" once it's uploaded to the cloud and Apple attempts to decrypt it using their secret key.

Google uploads a raw photo, which itself is essentially a meaningless value in the context of identifying CSAM, and Google "interprets" it on the server by hashing it and comparing it against some database.

In both cases, the values that are uploaded by the respective companies' devices don't mean anything, in the context of CSAM identification, until they are interpreted on the server.

I have a question about monopolies and market abuses.

Apple released their phone in 2007, the App Store in 2008, and in-app payments in 2009. During that time their marketshare was fairly small, and it didn't start to really grow until they expanded availability to the Verizon network in 2011.

Right at launch of the App Store, Apple announced its sales commission would be 30%. Then they extended that same fee to in-app purchases a year later. At the same time they set the rules that third-party app stores were not allowed, and that third-party payment processors could not be used.

I'm mentioning all of this history to make this point: Apple made these rules when they were not a monopoly by any definition. They released these products, with these rules, into a free market and let the market (both users and developers) decide which products to use and which products to develop for.

Now, obviously, between 2007 and 2021 the iPhone has been a wild success. Its platform has grown in users and developers every year.

So in terms of the framing of "market abuse", at what point between the launch of these rules and now did Apple cross that threshold between free-market competitor who can legally control their own platform to monopolist abusing its power?

I'm asking this question not just to make a point, but because I think it will be instructive for future companies to understand where in the growth curve the rules they started with can potentially cross over into being "abusive".

> "It doesn't account for... what sort of issues the person was having on that day specifically"

I'm not clear on what you are suggesting here. Are you saying they should contact people multiple times to make sure they weren't having a bad day the first time around?

2-3 days is common because any longer than that and the survey results will no longer be reflective of a specific point in time.

Also, if you check their crosstabs they have breakdowns for gender and age.

Having a poll in the field for two days is pretty common, and n=258 is definitely enough for a representative sample.

Right, but that sounds like the bigger issue here is that the model might spit out copyrighted material, not just that it scrapes it. The former seems like a technology problem that Microsoft can solve.

ML novice question: is this atypical when training models? Wasn't GPT-3 trained on a lot of copyrighted data? My gut instinct, which is based on very low-information, is that it would be pretty hard to train models if you could only use open-licensed material.

You say "Apple did nothing", but the only way a company would be paying a 30% commission to Apple is if the user's initial sign-up came through their platform.

In that situation, maybe the user would have eventually signed up through a different platform, and the company could have been directly paid 100% of the price. But it's also possible that the user would simply never have signed up, in which case Apple's 30% cut would look like a steal compared to the alternative (receiving nothing).

My 100% selfish take on this is that Apple's in-app payment flow is a pretty incredible user experience (double-tap a button + FaceID scan, and you're done), whereas being directed to a website and having to manually enter card details is a legitimate pain. I haven't pulled my credit card out of my wallet to enter my details on a phone in a long time. Haven't had to type my billing address, haven't had to find my CVC number. It's been nice!

I'm sure Apple is very pleased with the profit margins they get on in-app payments, but I'm equally certain that Apple's desire to keep people using their payment system is just as much about maintaining a great user experience for users.

Obviously the flip side is that some companies - Netflix! - then decide just not to offer payment in-app at all. That's also bad! But I don't know how to compromise on allowing apps to direct users to payment systems outside the app store without opening the floodgates to a million different payment experiences, which I (again, selfishly) would love to avoid.

One minor point of differentiation, it looks like while this uses screenshots to capture each step (correct me if I'm wrong), Cypress captures the DOM structure. You can go back to any step an inspect the raw code on the page, which offers a bit more context when debugging.

In all seriousness, if that's true you should do it and productize it. Based on the comments here it seems like many people prefer other browser scripting tools, so if you can replicate the novel features of Cypress in a better package, you could make money. Cypress raised almost $10M in their Series A.

For anyone new to Cypress, there are two things that I consider killer features:

1. It saves a snapshot of the DOM state before/after each test step. If you have a long acceptance test where it's deeply navigating through your app (i.e. visit "/admin", click on "Login", type "username", click "Submit"), using the GUI test runner will show each individual step in a side-panel. From there, you can hover over a step to see a snapshot of what the page looked like before that step was executed, and a snapshot of what it looked like after.

I had previously only used test runners where, when it would run the acceptance test, you would see a series of really quick flashes of the runner executing a bunch of steps faster than my eye could track, and then a big FAILED message. What failed? How did it fail? Cypress solves this problem by saving the DOM state and allowing me to traverse backwards through the state of the app until I find where things started to go wrong.

2. The cli test runner records a video of its test runs. Super useful in CI where you don't have access to the frontend. Cypress will record its run to an MP4 and you can investigate failures after-the-fact. This has solved a lot of "it works on my machine but not in CI" problems that seem to come up pretty often.

The other positive thing I have to say about Cypress is that I've actually found it more helpful as a development sidekick tool than as a test runner. Particularly when developing features on the front-end that require a lot of user interaction, many times it's easier for me to whip up a quick Cypress test that automates the user interaction (with assertions along the way to make sure the app is working) and then use the GUI Test Runner to do all my debugging.

It's not perfect and I definitely have some frustrations with it, but on the whole it's helped my productivity quite a bit.

In a previous HN post [1] where someone wrote up how to use B2 as an image host, the Cloudflare CEO chimed in and addressed rule 2.8 specifically, saying if Cloudflare workers are used for URL prettifying and redirecting, a different ToS is applied and that use-case would be fine.

Does that mean your video use-case would also be fine? I have no idea. An HN comment from the CEO doesn't seem like it would hold up if Cloudflare suddenly shut down your free account.

I'd love for Cloudfront to officially clarify the limits of the Cloudfront/B2 alliance in terms of external traffic. The confounding issue here is that B2, as a storage service, is not really intended for "serving web pages and websites" — it's for larger files, binaries, etc. — and therefore any traffic from B2 going through Cloudfront is sort of de facto in violation of 2.8.

[1]: https://news.ycombinator.com/item?id=20790857