The key point here is that data's security isn't enforced by the OS, but rather by the encryption algorithm itself. A non-OS forensics terminal would see the raw contents of the hard drive, which is to say, blocks of random-looking bytes that might be encrypted data or might just be truly random; without the corresponding password, there is no way to to tell.

Without any password (and assuming a robust encryption implementation), that's as far as they'd get. With a password, they'd be able to decrypt the blocks associated with that password and view the files for that account, but not any of the other blocks/accounts.

At that point, the FBI's nerds might well suspect that there are other accounts on the phone that they don't know about, but short of coercing the phone's owner into providing additional passwords, that's as far as they could get.

Joke's on you! I would definitely fly to a cheap third world country (like Madagascar) to get the surgery done using old school surgeons who have already practiced a thousand times.

You're assuming public web rings.

The way things seem to be going is that sharing information in the public sphere is being abused by the social media and AI companies, both by copying without attribution or reciprocal agreements, and by slop generation which drowns the information in noise. Also known as :Tragedy Of The Commons

If we assume private web rings are to be the future, then a lot of the above issues become moot as long as the access controls prevent the general public / AI companies from being part of the web ring. Sure once in a while the access will be breached, but mostly that might do the job required, which is: To allow timely and useful information sharing exclusively among the minority who belong to the private web ring.

What if someone took your photo without your consent, and uploaded it claiming you're one of their "friends"? Still fair game? No.

You people who claim that these predatory anti-privacy practices are all self-inflicted have no answer to the true problem: Most of the information available "publicly" and shared "freely" on the internet and social media sites is not actually "available" and is not "shared" by anyone who owns the information.

There's no putting the genie of stolen data in the bottle, you're right about that. But there is a strong case for government regulation world-wide, and forcing companies to *prove* every time that all the information they make use of for any purpose was legally acquired from the actual information owners.

Is this a heavy burden compared to the free for all wink wink going on now? Sure. Will that destroy a bunch of rich companies that temporarily prop up the US economy? Sure. Should it still be done regardless? Sure.

The goal would be to have a device OS with no concept of "correct blocks", i.e. there are only a bunch of blocks, all equally correct, and the OS uses some subset of them based on the password that was entered. Ideally the OS wouldn't even have the ability to tell anyone how many viable subsets there are.

Of course, "more cost-effective than SLS" is a pretty low bar. The actual goal would be "more cost effective than Falcon 9".

To me, it sounds like "well, we thought we had it isolated..."

There's nothing like saying "it can't be done" to make a nerd's imagination fire...

I'd start with a filesystem that stores all data encrypted, so that without the proper decryption key, there's no way to tell whether a given block of storage contains encrypted data or random noise. We'd overwrite the entire drive with random noise on install, of course.

Then on top of that, add a layer that determines which subset of the physical blocks to use based on a hash of the decryption key; so if you log in with password A you get a particular filesystem, but with password B you get a completely different/unrelated filesystem, and so on. Repeat as necessary for as many accounts as you like; all accounts which are equally real/valid; the only downside is additional drive space required.

The tricky bit would be coming up with an algorithm that tells each account which storage-blocks it is allowed to overwrite, but doesn't simultaneously provide a hint to the police that those other not-to-be-written-to blocks might actually contain hidden data rather than just noise. Well, that and the fact that just having this software installed would look suspicious, if they were able to recognize that it was present :)

Very much so! Unfortunately, they're still vaporware.

Or more accurately, I'd very much like a $10k BYD, but Detroit has bought DC wholesale to block any meaningful competition for Ford's Coal Roller 150.

That puts you on an iPhone 7 at the highest. iOS 15 is the oldest version installable on those, and not by coincidence, also the oldest iOS still receiving security updates. It was released five years ago as of September 20th.

So while you're 100% technically right... You might want to start planning for a new phone before October.

On the whole you're right because we simply don't know how to do it yet.

The GP isn't entirely wrong, though - It's also safe to say more training data is needed. Model weights vs training data follow a pretty clear power law, and current top models are already orders of magnitude underdetermined for their number of weights. Problem being, the current data set being used by the major players is effectively the sum total of all digitally-available human knowledge.

Unsupervised training is the hot new thing, but it doesn't really matter much if there's no more data for the AIs to find on their own.

Anthropic just accepted a USD$1.5 Billion court settlement for using illegally-acquired copyrighted content as training data.

Which side is not cheating here?

Ok, but some of us actually want to do something useful instead of just try to win the trophy of making a model run on the cheapest hardware. Unless you're generating at least ~30 tokens / second, it's not useful.

If you want to split the cost evenly, sure. But a team of 3 people working at a software company today typically pays the AI companies about $5,000-$10,000 a month, so I don't think they need to rely on the parts of the world where $400 is twice their income.

Bluntly, they can't. That genie is already out of the box for anything already available; and even going forward - Unless China itself blocks export, there's no realistic way to stop us from using a non-Chinese non-US mirror to grab them.

Realistically, hardware is the sole limiting factor for open-weight models, so restricting H200's is probably the closest we can come to a meaningful embargo. There's no practical way for most of us to run a model like Kimi K3 with almost 600GB (at BF16) of weights. Even using CPU offloading and/or multi-GPU sharding (if those are even possible for K3, I haven't looked at it in depth yet), the performance drop would give a throughput best measured in minutes or even hours per token, not particularly useful.

What this will accomplish is exactly what the OP quoted - It won't be possible to run these legally, so virtually all commercial use would be DOA. Joe Hobbyist couldn't care less though.