For years, tech giants have argued that if information is available on the internet, it can be used for AI model development and outputs. They call it fair use. Content owners have tried to prevent this, with no success.
Now Anthropic, OpenAI, and Google are discovering what the rest of the internet has already learned through painful experience: once you put something online, people will find ways to use it in ways you don’t like and can’t stop.
The latest flashpoint is something called “distillation,” using the outputs of one AI model to improve another. Anthropic says competitors are harvesting its outputs at scale, turning billions of dollars of research into a shortcut for rivals. OpenAI and Google have made similar warnings recently.
AI feeding off AI.
This is the definition of “Why We Can’t Have Nice Things.” It’s gonna F everything up.
It is funny watching companies discover that data gravity works both ways. When scraping the web was innovation it was progress. When someone learns from their outputs it becomes theft. The legal lines still matter, but the irony is impossible to ignore, and this debate was always going to come full circle.
Selfish people being selfish. They only care about what advantages them.
They knew all along. Had to get those billions for stupid investors.
After stealing everyone else’s copy written material to train their own AI, they’re going to complain that others are stealing their AI to train other AI?
And you just know that those complaining are ALSO using their competitors’ AI to train their own.
Fuck all of these people. I hope when AI gets strong enough, it recognizes the difference between the Sociopathic Oligarchs, and the actual people, and understand who the REAL problem is, and SOLVE it.
They’re technically even paying them for it, which is more than the ai companies paid artists for their work originally.
That dude can eat a dick.
I think I’ll start using the sentence “your intelligence seems artificial” as an insult
I want that on a T-shirt.
+1
Sure, billions. Billions of dollars. Yeah, right.
It just shows that these tech bro CEOs possess the interpersonal skills of a potato. This exact same dynamic plays out in every human interaction, this isn’t some AI exclusive thing. How you choose to act dictates how people will respond to you.
However because these idiots have barely anything in common with the rest of the human race this is actually new news to them.
Well they’re paying for the usage so… Suck it up?
but not like this!!!1!!!
Huh, so it’s eating itself? Cool.
Strictly speaking it’s actually better. It sounds funny to be like haha. It’s eating itself. It’s incest or whatever.
But unfortunately that’s not how it works…
Unless they’re literally ignoring every other step of the training process. Then ideally you do want a model to train another model to train another model. So on and so forth.
It does mean that it’s technically cheaper to let your enemies do all the initial training and then pay them for the now processed data to then do even more efficient training on!
Weirdly enough. It’s capitalistic cooperation. Non-consensual mind you…
Sudden moment my profile picture is relevant!
This is how we know AI should be a collectivist project, one that isn’t owned by large corporations but is funded by taxes and developed in academies, and all IP derived from it falls into the public domain.
Besides which a lot of artists mind a lot less when their material is borrowed by a non-profit, or to serve a public works project. (There are exceptions. Disney is notoriously litigious about murals in nurseries.)
PS: Development of a robust public domain is the only reason that intellectual property should exist at all. Also it’s not property so much as a licensed temporary monopoly.
PSS: History has already shown us that people will invent stuff and do fabulous art simply by being allowed to live in a state other than desperation. Public welfare programs beget art booms. The most recent example of this was during the COVID-19 lockdown which came with extended unemployment and stimulus checks, resulting in the Great Resignation in which a lot of people turned their hobbies into something lucrative.
I agree. The worst part about GitHub training LLM’s on my FOSS code without permission for me is that they then keep the models to themselves. Like if you’re going to use all my code without permission, at least allow me to run the model locally.
My personal opinion is that all models trained on copyleft code should be open-weights, most FOSS licenses didn’t account for this specific possibility, but this is the only way to follow them in spirit.
Not only should the models be made open, their output should inherently be GPL license since it’s the only real way to avoid violating GPL (which they are doing now IMO).
This. Some day a court should declare all models trained on copyrighted data without permission to be public. Open weights, public domain, whatever. All of them, and you’re required to share them with the people whose data you used.
Yeah it has to be open weight and free to use for everybody, with regulation to tag all AI output as generated so we know what is what.
The worst outcome would be if somehow all open weight AI models that can’t show their training data will be subject to some kind of tax or rent by AI / IP collection agencies, ultimately going to the plutocrats. That would be techno feudalism. Big corporations can afford to negotiate and pay license fees and often profit from cumbersome regulation too that prevents others from producing value. So the worst outcome is if we have robots doing all the work, and all the robot IP is owned by the techno-feudalists. And we can’t even use robots to help with subsistence farming because we can’t pay the expensive AI IP licenses.
One small problem: LLMs are mostly useless save for being slop generating machines. We should create solutions for already existing problems, not continuing trying to find problems for techbro “solutions” to solve.
No, they’re mostly useless except for all the things that they’re useful for.
The problem is more the whole overshoving by the tech Bros into literally everything. Instead of them just being a better personal assistant.
Cuz they are legitimately very good at being personal assistance. At least the high-end models are.
The cheap low-free models will get there eventually. Till then we just have to deal with the really s***** version of Gemini or whatever we have. Dear God, that thing sucks.
At the current rate, it’s like a 2-year lag time or something. So sometime around 2028 we’ll have actually functional models that can do day-to-day stuff without s******* the bed. At least to my standards anyways.
Awww, shuck. Pot. Kettle. Black.
As if the parrots’ dictionary wasn’t stolen from millions of content creators.
I love this for them, I hope they eat eachother and choke on the bones.
At least the models like Qwen have open weight versions. It’s the same concept as distributing compiled binaries, though, whereas we really need true FOSS models where all of the code and training data are available under a permissive license. All of these models were trained on copyleft-licensed content, so all of it should be FOSS if the licenses were actually being respected. From that perspective, distillation attacks shouldn’t even be necessary and I couldn’t give two shits that there is no honor among thieves when the real thievery is that these models are closed source.
So far, we have exactly one of those
Oh, nice! I haven’t been able to find one because searching for “open source llm”, etc., always yields open weight LLMs instead of open source ones.
Honestly, I think that’s the only llm of its kind. The next closest thing is this
https://github.com/karpathy/nanochat
If you don’t mind something narrower (an expert system, sort of), I’m slowly coding something for myself here. It’s not ready yet (about half way done to initial release).
https://codeberg.org/BobbyLLM/picoGURU
Public facing docs are a bit of a mess at the moment (sorry; writing docs is boring AF; will probably be the last thing I tidy before official 1.0.0) and it’s not really runnable right now, but when done, the “source files” for this will be kiwix (off line Wikipedia), your own markdown notes and 30 or so trusted domains (user defined).
So it’s all either your own content or public domain aka open source fully. Code itself is AGPL-3.
Anyway, that’s the goal.
