🔎 If you see an alert for analytics.usips.org, this is a self-hosted system I am running as a test. It is GDPR compliant and collects no invasive information.
Want to keep track of this thread?
Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading. Create account
However these AI lawsuits shake out, everyone who produces content for a living will add TOS to the website stating their content can't be used to train AI. Everyone who offers any kind of free service will hide in the EULA that everything you give it will be used to train AI. I fully expect Google to start feeding everyone's emails and photos into AI training sets sooner or later. Meta already uses all your public photos to train, going all the way back to 2007. It's a matter of time before it decides your private photos are fair game, too.
A site TOS may not be binding though (and I really dislike when they try to rule out Fair Use). But I'm pretty sure that hacking to get around paywalls isn't really legal...
Back when people could make respectable money from a couple of unobtrusive text-only Google Adsense banners on their blogs. Unfortunately bandwidth speeds were always going to increase and the spectre of the phone poster was already on the horizon.
The unit economics of cloud AI are very different from traditional Software as a Service. If you're running, I dunno, a forum where people can use it for free or pay to get a special banner, additional reaction stickers and a private board, each individual user (free or paid) isn't going to add anything noticeable to your running costs unless they're DoSing the site or something (and then you ban them for abuse). So free and premium features can be kind of arbitrary and a premium user doesn't really have to cost you more than a free user, depending how you structure things. Your running costs are mostly overheads rather than unit costs and additional users are marginal. Each user action is also marginal, so heavy users don't cost much more than inactive users. So a few premium users can effectively subsidize a much larger number of free users.
With an AI generation service, every time any user uses the core functionality of your website, there is an exact dollar charge on the API you're wrapping or the cloud GPU minutes you're using. If you have your own hardware the cost is less direct, but you can still only run so many generations at a time and the limits bite far quicker than they would on a traditional service because each user is doing something computation-heavy rather than just reading and writing a database. With traditional SaaS you're slapping an entrance charge on a bar, but with AI generation you're selling drinks.
So maybe you can afford to give away a few free generations to entice people to pay for more. In any case, free users are a calculated loss and premium users are an ongoing expense. Heavier users cost you more in a very legible way, hence most of these platforms have usage limits and credit systems. This is very different from traditional SaaS, where the cost of each additional user is usually marginal, doesn't change much based on their usage of the platform, and premium users don't cost significantly more than free users.
If you've got tons of money to burn and you want user data to train your own models, like OpenAI, Anthropic, Google (who also owns their hardware), etc, you can afford to take the hit to a greater or lesser extent (notice how Anthropic has very severe free tier limits vs how Google runs inference on free searches). But smaller platforms that don't train their own models don't really have much incentive to let you spend their money for free.
A site TOS may not be binding though (and I really dislike when they try to rule out Fair Use). But I'm pretty sure that hacking to get around paywalls isn't really legal...
Many paywalls can be defeated by disabling JavaScript, or visiting the article from a search engine instead of directly. It's hardly what I'd call hacking, though the courts may have a different view.
I doubt it is illegal to access a site with JavaShit turned off or blocked though. If one's browser is configured to not run JS, whatever loads from a site is what is there.
Making stuff under a paywall that needs JS to work is just incompetence.
against the TOS of the websites, often finding exploits to get around paywalls. Web scraping like they did is illegal, [...] produces content for a living
I see where you are coming from. The typical GenX geezer whining lines. My opinion in that regard might not be da law, but practically speaking, if one puts any data on the internet, without making it technically impossible to access it without ones explicit consent, which includes of course putting ones family photos on someones platform or chosing access guards that every 5 year old can circumvent, its going to be "stolen", used and if possible monetized; its digital data it can be copy pasted ad infinitum, in some way its worthless, it has no scarcity. No amount of paper law that is hardly able to be enforced is going to prevent that. And one might go so far that any law that attempts to prevent and disincentives third parties to utilize data on the internet produces more costs then the data shitted on the net is worth.
Putting data on the internet and thinking a TOS is an adequate protection is like putting a bike without a lock into the middle of an inner city and thinking a piece of paper that reads "please don't steal" is going prevent it being stolen, and then straining the limited resources of a juridical system whining about your own fucking retardation.
Character.AI is going bankrupt, and Disney it looking to aquire some of their data and tech. If anyone ever wanted to ERP with Donald Duck, you might be in luck soon.
Of course its fair game, its done daily at every millisecond on the scale of Billions of instances if automated access from Santa Fe, over Paris to Moscow and Jakarta. A few singular cases, hell even tens of thousands that might get paraded through the "public" legal shaming arena won't even make a dent, let alone coerce anybody into self regulating themselves according to the retarded imagination of GenX pseudo moral and legal code when it comes to on how to treat data on the internet.
Character.AI is going bankrupt, and Disney it looking to aquire some of their data and tech. If anyone ever wanted to ERP with Donald Duck, you might be in luck soon.
You forgot the part where Disney hired the Jeet running it as their CTO as well. The looting of America continues at breakneck speeds can't wait for all of the new kingdom hearts characters to be indian.
A site TOS may not be binding though (and I really dislike when they try to rule out Fair Use). But I'm pretty sure that hacking to get around paywalls isn't really legal...
Restricting the usage & distribution of content is usually binding. A court could rule that merely putting it on the internet means AI companies can embed it in model weights, overriding any T&Cs you try to impose.
The unit economics of cloud AI are very different from traditional Software as a Service. If you're running, I dunno, a forum where people can use it for free or pay to get a special banner, additional reaction stickers and a private board, each individual user (free or paid) isn't going to add anything noticeable to your running costs unless they're DoSing the site or something (and then you ban them for abuse)
Bandwidth costs money. If you have 5 users, sure, it's negligible, but if you have 50,000, it costs real money. AI scraper traffic is already hitting websites so hard that it's incurring significant costs. Current AI economics require everyone else to pay to create content, then pay to host the content, then give it to the AI companies for free, who then turn around and charge you for the content.
This isn't sustainable, and it obviously isn't sustainable. If people can't make money from creating new content, they just won't. AI already has a problem of the content it scraping being mainly AI-created, driving the models to collapse around sloppified averages, and this will only get worse as they put all the genuine content creation engines out of business.
I find it supremely ironic how China, despite having the most totalitarian collectivist bugman regime to ever exist, provides more support to local and open source LLMs compared to America despite it's relatively individualist culture
Anyway does anyone want to try NiggerLLM for their next projects?
In hindsight it honestly wasn't all that golden. We didn't know then but we signed our freedoms away by signing up for free and convenient services, allowing many of the worst of the worst to become what they are now. I now pay for internet services like email and with that vote with my wallet and while it kinda feels like trying to stop a tsunami with a mop, I can only encourage others to do the same.
Since this is like the third thread I'm looking at regularly that slowly morphed into a place in which people post How They're Totally Not Mad about AI maybe really somebody ought to make some AI debate thread so people can keep circlejerking how this is just a fad that'll go away, you know, just like computers or the internet, while others can write their fanfic about the voice of world control who'll bring them UBI.
I find it supremely ironic how China, despite having the most totalitarian collectivist bugman regime to ever exist, provides more support to local and open source LLMs compared to America despite it's relatively individualist culture
China is ideologically communist. It's only natural for them to believe that everyone should have access to these revolutionary tools, and not only the 1%.
Since this is like the third thread I'm looking at regularly that slowly morphed into a place in which people post How They're Totally Not Mad about AI maybe really somebody ought to make some AI debate thread so people can keep circlejerking how this is just a fad that'll go away, you know, just like computers or the internet, while others can write their fanfic about the voice of world control who'll bring them UBI.
I don't think anyone is arguing about if this is just a fad at this point. LLMs are too big and useful for that right now.
There are really 4 problems which this thread is trying to solve: discussion on the tech itself, reactions to its usage, the current state of the industry, and the economic consequences and folly that comes from investing in it.
There are sub threads for various technologies, we have a stable diffusion thread, but we don't have a LLM or agentic coding thread. Ideally this thread should be about the social and economic aspects, and a more specific thread about LLMs and agentic coding focused on the tech made. They should be linked in the OP. The debates at this point appear to be mostly technical and related to the way LLMs function, so you would get rid of a lot of that from this thread by moving them to a specific LLM thread. If that matters.
This is not unlike dumping, where the Chinese government heavily subsidizes something, the dumps it on a Western market at a loss to bankrupt local industry. Maybe it's not "communist," but it's not unprecedented.
Nvidia is releasing a new MoE, Nemotron Bolt 3.5 74B A9B tomorrow. This includes Q2 quants with quantization aware training that supposedly fit in 24gb of VRAM. It would be neat if other model makers hop on board that trend.
“There's a fair amount of [disaster] prepping and bio-bunker purchasing going on” among these researchers, says Edouard Harris, Gladstone’s chief technology officer.
This basically means there's a possibility to put a neural net encrypted inside a model that could could be triggered externally, without being able to tell what it is or does without being able to unlock it.
