I'm strictly speaking never going to think of model distillation as "stealing." It goes against the spirit of scientific research, and besides every tech company has lost my permission to define what I think of as theft forever
replies(3):
Though it's poetic justice that OpenAI is complaining about someone else playing fast and loose with copyright rules.
Even if you give GenAI unlimited time, it will not develop its own writing/drawing/painting style or come up with a novel idea, because strictly by how it works it can only create „new” work by interpolating its dataset
There is no evidence whatsoever to support that humans create "new, spontaneous thoughts" in any materially, qualitatively different way than an AI. In other words: As a Turing-computable function over the current state. It may be that current AI's can't, but the notion that there is some fundamental barrier is a hypothesis with no evidence to support it.
> Even if you give GenAI unlimited time, it will not develop its own writing/drawing/painting style or come up with a novel idea, because strictly by how it works it can only create „new” work by interpolating its dataset
If you know of any mechanism whereby humans can do anything qualitatively different, then you'd have the basis for a Nobel Prize-winning discovery. We know of no mechanism that could allow humans to exceed the Turing computability that AI models are limited to.
We don't even know how to formalize what it would mean to "come up with a novel idea" in the sense you appear to mean, as presumably, something purely random would not satisfy you, yet something purely Turing computable would also not do, but we don't know of any computable functions that are not Turing computable.
You need to be a bit more expansive. Turing-computable functions need to halt and return eventually. (And they need to be proven to halt.)
> We know of no mechanism that could allow humans to exceed the Turing computability that AI models are limited to.
Depends on which AI models you are talking about? When generating content, humans have access to vastly more computational resources than current AI models. To give a really silly example: as a human I can swirl some water around in a bucket and be inspired by the sight. A current AI model does not have the computational resources to simulate the bucket of water (nor does it have a robotic arm and a camera to interact with the real thing instead.)
The question of whether the mechanism of learning in a human brain and that in an artificial neural network is similar is a philosophical and perhaps technical one that is interesting, but not relevant to why intellectual property law was conceived: To economically incentivize human citizens to spend their time producing creative works. I don't actually think property law is a good way to do this. Nonetheless the question when massive capital investments are used to scrape artists' work in order to undercut their ability to make a living from that work for the benefit of private corporations that do not have their consent to do this is whether this should violate this artificial notion of intellectual property that we have constructed for this purpose, and in that sense, it's fairly obvious that the answer is yes
If you want to argue it's a distraction, argue that with the person I replied to, who was the person who changed the focus.
This is pedantry. Any non-halting function can be decomposed into a step function and a loop. What matters is that step function. But ignoring that, human existence halts, and so human thought processes can be treated as a singular function that halts.
> Depends on which AI models you are talking about? When generating content, humans have access to vastly more computational resources than current AI models. To give a really silly example: as a human I can swirl some water around in a bucket and be inspired by the sight. A current AI model does not have the computational resources to simulate the bucket of water (nor does it have a robotic arm and a camera to interact with the real thing instead.)
An AI model does not have computational resources. It's a bunch of numbers. The point is not the actual execution but theoretical computational power if unconstrained by execution environment.
The Church-Turing thesis also presupposes an unlimited amount of time and storage.
See https://scottaaronson.blog/?p=735 'Why Philosophers should care about Computational Complexity'
Basically, what the brain can do in reasonable amounts of time (eg polynomial time), computers can also do in polynomial time. To make it a thesis something like this might work: "no physically realisable computing machine (including the brain) can do more in polynomial time than BQP already allows" https://en.wikipedia.org/wiki/BQP
The argument I made in no way rests on a "complete picture of human learning". The only thing they rest on is lack of evidence of computation exceeding the Turing computable set. Finding evidence of such computation would upend physics, symbolic logic, maths. It'd be a finding that'd guarantee a Nobel Prize.
I gave the justification. It's a simple one, and it stands on its own. There is no known computable function that exceeds the Turing computable, and all Turing computable functions can be computed on any Turing complete system. Per the extended Church Turing thesis this includes any natural system given the limitations of known physics. In other words: Unless you can show knew, unknown physics, human brains are computers with the same limitations as any electronic computer, and the notion of "something new" arising from humans, other than as a computation over pre-existing state, in a way an electronic computer can't also do, is an entirely unsupportable hypothesis.
> and it needs to be addressed wherever possible that the ontological question is not what matters here
It may not be what matters to you, but to me the question you clearly would prefer to discuss is largely uninteresting.