I'm not saying that as an insult that's just a literal fact.

It is extraordinarily expensive to train a police officer. Much of this expense is because of corruption. Most of the training that we give them these days is pseudo-military training that looks cool to idiots. Think gravy seals. But we still have to pay for it all and the city has to pay for it.

The upshot to that is doing the actual job is incredibly boring. It's mostly pulling people over and giving them tickets. Something like 70% of police officers will never even draw their sidearm.

However the job pays reasonably well given how bad the economy is and has a clear career path to retirement. Along with a lot of opportunities to make extra money on the side.

So the problem cities had was people would go through all the training get bored out of their mind at the job and quit and go do something else. So they started to do testing to detect people or too smart. People who would get bored.

I haven't done it myself but I have friends that have applied to be a police officer and been turned down and when they looked into it it literally came down to they were too smart and the algorithm said they would quit the job after 6 months to a year haven't gotten all that expensive training...

As a result we intentionally select dumb people to be police. As an added bonus if you need them to bust heads at a union meeting they're all on board...

Dumb people tend to be violent because when you're not all that bright you quickly run out of options. Like the old saying goes, War is the continuation of political policy by other means...

So you get a lot of police officers beating their wives. You get a lot of cops shooting people they shouldn't be shooting. That sort of thing.

And you get a lot of cops spying on their exes.

They're a well known and well understood solutions to this problem. Other countries have done them. But America generally isn't up for that

No. We do not know how the decision was made. We can only look at logs and guess when patterns are found

Please stop commenting on things you don't understand. This is a conversation about attribution graphs, not CoT.

Unsoldering the ram chips and replacing them takes some effort and skill, and then you end up paying for 2 sets of ram chips - the old ones (which you cant do anything with) and the new ones. Given the effort and cost involved, this is only worth doing if:

1) the higher ram model is heavily marked up vs the incremental component cost so buying the ram yourself is significantly cheaper.
2) the same ram is genuinely expensive, but much cheaper ram is available and still works without noticeable side effects

If it's all about markup then it's pure profiteering and screwing the users.

If the only way to undercut the official models sufficiently to make the whole operation worth doing is by using inferior components then there is *some* legitimacy to trying to stop this, but the approach is wrong.
If someone is going to the effort of replacing the ram chips with something inferior to resell the unit then they're not going to think twice about shipping with an old bootloader either, or modifying the bootloader to bypass the check, or working out how to get the bootloader from the genuine larger ram models to work on the lower ram models etc.
Sure the hacked/old bootloader might get noticed later if someone tries to update it and bricks the device, but low quality ram or a poor soldering job could just as easily brick the device too. By the time this happens the seller will be long gone anyway.

This DOES hinder repairs. For instance you might have multiple faulty units with different faults so you take the working RAM from unit A which has a faulty CPU, and switch it onto unit B which had bad ram.

A good cop in a corrupt society is the same as a "good guy with a gun" in a society obsessed with guns and violence... such a tiny minority that they're pretty much non existent.

Cardinal Richelieu once a said, "If you give me six lines written by the hand of the most honest of men, I will find something in them which will hang him". Show me a cop in the US that's never taken a bribe, even just a sweetener for doing the right thing and I'll show you a liar. Even if just through the people who contribute to their pensions. Even if you can find someone who has managed to walk through that and some how remained completely clean (in the words of Londo Mollari "I can only assume you have not been paying attention") then back to what Richelieu said, he'll have still done something technically wrong even if it wasn't morally wrong. For example, the nice officer who let me off for speeding on the A4 near Heathrow a few years back with a bit of a tongue lashing, technically he broke the law as I was speeding and he should have wrote me up but he didn't as I wasn't a knob or driving like an idiot (and clearly the officer wasn't a knob either.. this is what happens when nice people meet).

And that's the UK, the US has long tolerated higher levels of corruption but in recent years that seems to be taken to truly herculean levels. With the egregious and obvious levels of corruption you're tolerating in your highest office, and there is no "I didn't know" excuse, you elected him twice... why would you expect the people with a little authority like beat cops to be goodly and honest when your entire country is no longer simply permissive of corruption, nor just expecting of it but openly supporting it.

(A) AI output has no copyright. There may be copyright from stolen training data still in place though if the AI produces too verbatim. The latter may beg worse with numerous pending lawsuits.

Indeed. And that gives at the absolute least gross negligence. That would make them liable for all damage done. But I suspect more is possible.

Hahaha, no. Microsoft, Google, etc. though that as well until the first really large fines came rolling in.

Well, we may get lucky as the manipulators may do to things on the cheap. Oligarchs may simply be incompetent as fascists. Will definitely be interesting to watch.

So, the reality is that the world "ran out of training data" for the most part years ago, and the models have gotten exponentially better relative to a number of parameters. Claude 3.7 Sonnet was released 1 1/2 years ago, and today it benchmarks about the same as Qwen 3.6 Sonnet 9B, a model two orders of magnitude smaller than it, and which is itself two generations out of date. And a large chunk of this is done with synthetic data - aka, data created by other models.

It's simply a myth that "data created by models consumed by other models makes them worse". In practice, it's used to make them vastly better. Models aren't collagers, they're reasoners. Learning the results of reasoning, the results of trial and error, etc helps build a stronger base for more advanced reasoning. Also, our training algorithms, while reaching a denser knowledge compression than human brains, are less efficient learners than human brains (per unit data), so they need to see the same sort of data from "many different angles", to substitute for our process of "mulling over" new information.

(Yes, it is possible to set up contrived scenarios where, say, an small image model is fed only its own outputs on loop, little bits of knowledge slowly being lost each go-round, in a situation equivalent to leaving a person alone with their thoughts in a dark room for ten thousand years - but even a tiny percent of new fresh data added to the mix prevents this degradation.)

And as for the article itself, they made it sound like they're talking about, say, programmers banned from using AI at OpenAI, but it's nothing of the sort. These are data labelers. In the old days, they used to be far more common, and in wide use in all types of model creation. That's no longer the case; they exist for special cases. For LLMs, this is much more limited:

* Subject matter experts: people with rare professional-tier knowledge. Often used to validate model outputs, where nobody else could (for example, OpenAI hires mathematicians to validate their models' proofs)

* Chain of thought / logic auditing. Increasingly important now that models are showing increasing signs of poor alignment. You can automate this a lot, but you really still do want a human in the loop *somewhere*, in case your auditors get compromised.

* Side by side comparative rating: Which model's output do you like more, A or B?

* Evaluating reported outputs where users reported that they thought the response they received was bad, and if there's actually anything wrong, copyediting the output for training.

* Adversarial prompt generation / jailbreaking and evaluation. Again, you *can* have models do this (and companies often do), but you don't want to just rely on them.

* Trying to set the bounds on whether given queries should be refused or not (for example, "How do explosive reactions happen in chemistry?")

Basically, a switch from "click work" to "knowledge work". This is no longer the era of "Write a poem about cats" or "Explain how to solve this algebra problem" to build up a training dataset. You're getting paid to think, not to repeat a rote task.

Other types of labelers aren't as far along. Multimodal data is less advanced than text, so you'll still for example have people labeling things in videos, transcribing heavily-accented audio, grading text-to-video consistency, things of that nature. Probably the least advanced field is robotics, so there's still an awful lot of manual evaluation and correction in that.

But anyway, if you're hired to do any of the above, it's because they specifically want you to do that. Having an AI model do the above (beyond the listed caveats) entirely defeats the purpose.

No. A black box is something you don't understand how it works. We can understand any and every decision that they make. We simply have to look.

I guess the only positive thing I can say about them is: despite how tough the models were to work with then and get good content out of them, these people were generally obsessive over their waifus and porn, so at least they weren't making like 8-fingered monstrosities. You could tell that they spent many hours zoomed in, upscale-regenerating and collaging and photoshopping over every imperfection.

hasn't really taken over human actresses despite that industry usually being in forefront of all kinds of tech, I wouldn't bet on this one either.

I dunno. This isn't exactly my space, and I haven't even been involved much in image generation at all in years, but in the old days at least, the Stable Diffusion Reddit and the model sites like Civitai were just *flooded* with stuff from horny guys, like 80-90% of the content. It was really annoying. Due to anti-porn restrictions, the Stable Diffusion Reddit mostly got flooded with waifus or similar, but Civitai was mostly porn models, or at least general models finetuned to allow porn. People were even widely using waifus and porn in their tutorials and user guides for image generation tools, even though the the tools were just for general image generation purposes. Oh, and YHVH protect you if you dared complain to any of the waifu or porn posters about any of this - you'd get a flood of "WHAT, ARE YOU A PRUDE???", "HAVEN'T YOU EVER SEEN A WOMAN BEFORE???" , etc and get modded to oblivion. And perhaps the worst part of it was just how damned derivative it all was. There was zero creativity over any of it, zero diversity of artistic style, zero attempt to do something aesthetically new, just pure hormone-driven churn, generally either straightforward-photographic style, anime style, or a Midjourney-ish digital art style.

So I rather have to disagree with this statement. Porn users are a major share of AI image generation, at least with run-it-yourself models (as most commercial models ban it).

I mean, you're not wrong ;)

It's like the whole "metaverse" thing of trying to reinvent shopping in a 3d gamelike environment. Even in video games it's common to not have to actually hunt for and pick items off the shelves in 3d, instead just choosing from a dialog. Just because something sounds shiny on paper doesn't actually mean it's good in the real world.

You don't need to see the person on the other side to chat with them. You don't even need to hear them in most cases - text is just fine, and actually helps prevent misunderstandings.