> Going to be interesting to see how human systems deal with this.
At least a bunch of lawyers already got hit when their court filings cited hallucinated cases. If this trend continues, I'll not be surprised when some end up disbarred.
"open training" is something that won't ever happen for large scale models. For one, probably everyone's training datasets include large amount of questionable material: copyrighted media first and foremost (court cases have shown that AI models can regurgitate entire books almost verbatim), but also AI slop contaminating the dataset, or on the extreme end CSAM - for Grok to know how the intimate bits of children look like (which is what was shown during the time anyone could prompt it with "show her in a bikini") it obviously has to have ingested CSAM during training.
And then, a ton of training still depends on human labor - even at $2/h in exploitative bodyshops in Kenya [1], that still adds up to a significant financial investment in training datasets. And image training datasets are expensive to train as well - Google's reCAPTCHA used millions of hours of humans classifying which squares contained objects like cars or motorcycles.
Source: EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS) has released Apertus, Switzerland’s first large-scale open, multilingual language model — a milestone in generative AI for transparency and diversity. Trained on 15 trillion tokens across more than 1,000 languages – 40% of the data is non-English – Apertus includes many languages that have so far been underrepresented in LLMs, such as Swiss German, Romansh, and many others. Apertus serves as a building block for developers and organizations for future applications such as chatbots, translation systems, or educational tools. The model is named Apertus – Latin for “open” – highlighting its distinctive feature: the entire development process, including its architecture, model weights, and training data and recipes, is openly accessible and fully documented.
I’m not convinced that Grok’s dataset must contain CSAM for it to generate CSAM. Surely a combination of nude adults and clothed children would allow for it to synthesize CSAM?
(Disclaimer: I’m not in favor of AI in general and definitely not in favor of what Grok is doing specifically. I’m just entirely sold on the claim that its dataset must contain CSAM, though I think it is probably likely that it has at least some, because cleaning up such a massive dataset carefully and thoroughly costs money that Elon wouldn’t want to spend.)
Agree that this makes it unlikely we see frontier training data OS'd but this is a separate problem from software and infrastructure transparency, which has none of those constraints. Training stack, the parallelism decisions, documented failure modes are engineering knowledge and there's no principled reason it doesn't ship.
The human labor aspect is very little discussed and essential and very abusive, I am sure.
People think of these models as "magic" and "science" but they do not realize the immense amount (in human years) of clicking yes/no in front of thousands of pairs of input/outputs.
I worked for some months as a Google Quality Rater (wow), and know the job. This must be much worse.
I agree full transparency on data adds several other challenges. Still, even releasing the software and infrastructure aspects would be a huge step from where we are now. Also, some recent work has shown pretraining filtering to be possible and beneficial which could help mitigate some concerns of sensitive data in the datasets.
> Which is to NOT use those platforms that require age verification.
That is getting harder and harder. Platforms that are not susceptible to age verification (yet?) are on their way out - when have you written an email the last time for personal (i.e. non-work, order or customer support related) reasons? A physical letter [1]? The (root) cause is, centralized platforms like Whatsapp are much much more convenient and on top of that network effects apply - when 90% of your social connections use Whatsapp exclusively, it's hard to not use Whatsapp as well.
And then you got digitalization of government services and banking. More and more governments push for the removal of paper forms and require a web service. Banking regulations enforce 2FA, which almost always comes in the form of a phone app. The web services require a browser and an OS, which may require age verification sooner than later (see the recent spat about California's law), and the phone apps are only available for the walled gardens of unrooted, Play Store certified Apple and Android phones - that can and will be forced to verify ages as well.
Hard cash is out as well, many governments have set hard caps on cash transactions due to "anti money laundering" laws, in other countries you need to have a bank account to pay for mandatory things like taxes or public broadcast fees [2], and an increasing number of vendors refuses to accept cash as well due to the associated handling cost and risk of fraud (i.e. employee theft) and robbery.
That last point alone will make it impossible to survive in society without engaging with one or more of the walled gardens.
And mercy be upon you if the US Government decides to put you on one of their black lists. No more banking, even as an European, because everything touches VISA/MC/SWIFT, your cloud accounts (and with it your phone and app stores), all gone, you are now an unperson [3].
> Many HNers strongly argue that it's absolutely impossible to distinguish between AI text and non-AI text.
And it's becoming more and more difficult - not just by AI getting "better" (and training removing many of the telltale signs), but also because regular people "learn" to write like an AI does. We're seeing it with "algospeak" - young terminally online people literally say stuff like "unalived" in the meatspace nowadays.
> It failed on Reddit because Reddit is maintained by a bunch of volunteers to whom Reddit provides woefully, woefully, horrifically underdeveloped tooling to automate their communities in a more nuanced way.
And on top of that, some of said "volunteers" are power-hungry, petty, useless fucking morons. Especially the large subreddits tend to be run by people I wouldn't trust to boil some pasta without triggering a fire alert, and yes I know people who manage that.
not entirely, IOMMU is a thing, that is IIRC how Amazon and other hyperscalers can promise you virtual machines whose memory cannot be touched even in the case the host is compromised (and, by extension, also if the feds arrive to v& your server).
>how Amazon and other hyperscalers can promise you virtual machines whose memory cannot be touched even in the case the host is compromised (and, by extension, also if the feds arrive to v& your server).
Even if we take those promises at face value, it practically doesn't mean much because every server still needs to handle reboots, which is when they can inject their evil code.
This is excellent. The ability to trick remote servers into believing our computers are "trusted" despite the fact we are in control will be a key capability in the future. We need stuff like this to maintain control over our computers.
Well, it kind of is actually. The previous iteration of the design didn't have that vulnerability but it was slower because managing IVs within the given constraints adds an additional layer of complexity. This is the pragmatic compromise so to speak.
Does it count as a conceptual problem when technical challenges without an acceptable solution block your goal?
Hosting tor outbound server at home is stupid idea.
Your home is gonna be raided by Police and you will wait months or year to get your shit back and then if nothing, gonna be charged for having pirated windows and Photoshop lol
These days, every American's threat model should include being v& by the feds, and here in Germany, the situation isn't much better, you can get v& for saying the Minister of Interior is a dick [1].
Yes, this was later on ruled unconstitutional, but it doesn't change the facts, and, worse, Germany doesn't have a "fruit of the forbidden tree" rule.
That's BS. Linux/OtherOS on the PS3 only existed so that Sony could evade European tariffs on pure gaming consoles that were markedly higher than on general-purpose computers. Once the tariffs went, so went OtherOS support.
In any case OtherOS didn't have access to the full system resources and, on top of that, Sony actually lost money on PS3s because it was priced as a loss-leader, with game purchases being supposed to earn the actual money.
> Imagine, for instance, if you bought a flat head screwdriver, but the manufacturer told you that you could never, ever, under any circumstances use it to pry something open. It was stricly to be used for installing or removing screws.
Try filing a warranty claim if it bends, the manufacturer will go and tell you to kick rocks.
That is fine, they're not required to support unexpected use cases. Not the same as forbidding you from using it as you see fit. It's simple really, they can do what they want with their resources, you can do what you want with yours, especially those you paid for.
But you only have a license to use the screwdriver, it's still their property. You aren't entitled to free use of someone else's property, of course. Just because it's in your possession doesn't mean it's yours!
(This is supposed to be satire but feels scarily accurate anyway.)
Everytime I use an Autodesk product (6days a week) I lament this court case.
The whole set of software IP rules seems like mental gymnastics to justify a career (that I like, support, and benefit from) rather than rules that come from axioms or ethics that make sense.
> A good guy vs bad guy narrative is easy to sell and distracts people from the fact that human life is not valued by any of them.
Iran slaughtered 30k people in a matter of days for the crime of "protesting". No tears shed for the Mullahs here, IMHO Israel and the US are doing the world a service here by finally cleaning up the last terrorist regime keeping the region in a constant state of aggression. Note that before and after Oct 7th, it only was Iranian backed forces stirring shit (Houthis, Hezbollah, Gaza's Hamas), while everyone else stayed put.
If you don’t think the current US administration would gladly slaughter 30k people in a matter of days for protesting if they thought they could could get away with it, you’re not paying attention.
Iran isn't being bombed because the leadership is nasty (terrorist regime). They are being bombed because they are s rival power to Greater Israel expansion, and they are a major oil exporter that doesn't use $US.
The leader of Syria is a Sunni terrorist but he stays in place because he blocks arms shipments through Syria from Iran to Hezbollah. That suits Israel. Israel doesn't care how the Syrian leader treats his own people so long as he continues to disallow arms to Hezbollah from Iran. And the USA goes along with whatever Israel wants.
It has nothing to do with how the Iranian regime treats its own people. Just look at how the Saudi regime treats its people.
> The leader of Syria is a Sunni terrorist but he stays in place because he blocks arms shipments through Syria from Iran to Hezbollah. That suits Israel. Israel doesn't care how the Syrian leader treats his own people so long as he continues to disallow arms to Hezbollah from Iran. And the USA goes along with whatever Israel wants.
That's one thing, the more important (and sad) thing is that the Islamists are the ones who eventually won the war and there is no one left as a contender for governing the country. The Kurds alone are too small.
In Iran the situation is different, as Iran always had a vibrant civil society that is only held back by the Mullahs' sheer military and police gun power. I'm confident they will manage something decent once Israel and the US have bombed enough of the Mullahs, IRGC, Basij etc. to cause the rest of them to flee to Moscow.
Meanwhile the US is lifting sanctions on Russian oil while Russia bombs Ukraine. Turns out peace or democracy is not what they care about, it is regional dominance for Israel.
What has blown my mind is how surprised people seem to be by all of this, it's like, they never imagined these people were capable of doing this...remember when it was "just jokes..." ?
At least a bunch of lawyers already got hit when their court filings cited hallucinated cases. If this trend continues, I'll not be surprised when some end up disbarred.
reply