5 ms·

> The fact is that “Probabilistic plagiarism” is a mechanical process, so as much as you might like to anthropomorphize it for the sake of your argument I did

> The fact is that “Probabilistic plagiarism” is a mechanical process, so as much as you might like to anthropomorphize it for the sake of your argument I did not anthropomorphize anything. “Learning” is the proper term. It takes input and applies it intelligently to future tasks. Machines can learn, machine learning has been around for decades. Learning doesn’t require biology. My statement is that it is not plagiarism in any form. There is no claim that the content was originally authored by the LLM. An LLM can learn from a textbook and teach the content, and it will do so without plagiarism. Just as a human can learn from a textbook and teach. Making an analogy to a human doesn’t require anthropomorphism.

neuroelectron 2y ago
> I did not anthropomorphize anything. Machines don't learn. They encode, compress and record.
- Earw0rm 2y ago
  Until or unless the law decides otherwise. The 2020s ethic of "copying any work is fair game as long as you call the copying process AI" is the polar and equally absurd opposite to the 1990s ethic of "measurement and usage of any point or dimension of a work, no matter how trivial, constitutes a copyright infringement".
- wiseowise 2y ago
  Semantics.
- ben_w 2y ago
  It's been a term of art since 1959, and the title of a research journal since 1989: https://en.wikipedia.org/wiki/Machine_Learning_(journal) https://en.wikipedia.org/wiki/Machine_Learning_(journal)
  - neuroelectron 2y ago
    Does a research journal on aliens prove aliens exist?
    - ben_w 2y ago
      As this is a semantics debate, their actual existence is irrelevant. A research journal on extra-terrestrial aliens would prove that the word "aliens" is used to mean "extra-terrestrials" and that the word doesn't just mean "foreigners": https://www.law.cornell.edu/uscode/text/8/chapter-12/subchapter-II/part-II https://www.law.cornell.edu/uscode/text/8/chapter-12/subchap...
      - fuzzfactor 2y ago
        >“Learning” is the proper term. It takes input and applies it intelligently to future tasks. To me "learning" is loading up the memory. "Thinking" is more like applying it intelligently, which is not exactly the same, plus it's a subsequent phase. Or at least a dependent one with latency. >Machines can learn, machine learning has been around for decades. Learning doesn’t require biology. Now all this sounds really straightforward. >Machines don't learn. They encode, compress and record. I can agree with this too, people are lucky they have more than kilobytes to work with or they'd be compressing like there's no tomorrow. But regardless, eventually the memory fills up or the data runs out and then you have to do something with it, whether very intelligent or not. Might as well anthropomorphize my dang self. If you know who Kelly Bundy is, she's a sitcom character representing a young student of limited intellect and academic interest. Part of the schtick was that she verbally reported the error message when her "brain is full" ;) It was just an observation, no thinking was required or implied ;) If the closest a machine is going to come is when its memory is filled, so be it. What more can you expect anyway from a mere machine during the fundamental data input process? If that's the nearest you're going to get to organic learning, that'll have to serve as "machine learning" until more sensible nuance comes along. Memory can surely be filled more intelligently sometimes than others, which should make a huge difference in how intelligently the data can handled afterward, plus some data is bound to be dramatically more useful than others too. But the real intelligent stuff is supposed to be the processing done with this data in an open-ended way after all those finite elements have been stored electronically. To me the "learning" is the data input, or "student" phase, and the intelligence is what you do with those "learnings" if it can be made smart. It can be good to build differing scenarios from the exact same data, and ideally become better at decision-making through time without having more raw data come in. Improvements like this would be the next level of learning so now you've got more than just the initial data-filling. As long as there's room in the memory, otherwise you're going to have to "forget" something first when your brain is already full :) I just don't think things are ideal myself. >journal on extra-terrestrial aliens would prove that the word "aliens" is used to mean "extra-terrestrials" Exactly, it proves that the terminology exists, and gives it more meaning sometimes. The aliens don't have to be as real as you would like, or even exist, nor the intelligence.
tsimionescu 2y ago
The law on copyright doesn't depend on the word "learning", it depends on whether it's a human doing it, or a mechanical process. If a human reads a book and produces a different book that's sort-of-derivative but doesn't copy too many elements too directly, then that book is a new creative work and doesn't infringe on the copyright of the original author. For example, 50 Shades of Gray is somewhat derivative of Twilight (famously starting as a Twilight fan-fic) but it's legally a separate copyright. Conversely, if you use a machine to produce the same book, taking only copyrighted text as input and an algorithm that replaces certain words and phrases and adds certain passages, then the result is a derivative work of the original and it infringes the copyright of the original author. So again, the facts of the law are pretty simple, at the moment at least: even if a machine and a human do the exact same thing, it's still different from a legal perspective.
- DoctorOetker 2y ago
  machines work on behalf of humans, any human could similarly claim their typewriter did it
  - tsimionescu 2y ago
    That's still irrelevant. The distinction the law makes is between human creativity vs machine processes. A human may or may not have added their own creativity to the mix, and the law falls on the side of assuming creativity from humans (though if there is proof that the human followed some mechanical process, that likely changes things). Machines are not creative, by definition in the law today, so any transformation that a machine makes is irrelevant, the work is fundamentally the same work that was used as input from a legal perspective. That this work is done on behalf of a human changes nothing, the problem the law has with this is that the human is copying the original work without copyright, even if the human used a machine to produce an altered copy. Whether they used bcrypt, zip, an mp3 encoder, a bad copy machine, or a machine learning algorithm, the result is the same: the output of a purely mechanical process is still a copy of the original works from a copyright perspective.

neuroelectron 2y ago

> I did not anthropomorphize anything. Machines don't learn. They encode, compress and record.

Earw0rm 2y ago
Until or unless the law decides otherwise. The 2020s ethic of "copying any work is fair game as long as you call the copying process AI" is the polar and equally absurd opposite to the 1990s ethic of "measurement and usage of any point or dimension of a work, no matter how trivial, constitutes a copyright infringement".
wiseowise 2y ago
Semantics.
ben_w 2y ago
It's been a term of art since 1959, and the title of a research journal since 1989: https://en.wikipedia.org/wiki/Machine_Learning_(journal) https://en.wikipedia.org/wiki/Machine_Learning_(journal)
- neuroelectron 2y ago
  Does a research journal on aliens prove aliens exist?
  - ben_w 2y ago
    As this is a semantics debate, their actual existence is irrelevant. A research journal on extra-terrestrial aliens would prove that the word "aliens" is used to mean "extra-terrestrials" and that the word doesn't just mean "foreigners": https://www.law.cornell.edu/uscode/text/8/chapter-12/subchapter-II/part-II https://www.law.cornell.edu/uscode/text/8/chapter-12/subchap...
    - fuzzfactor 2y ago
      >“Learning” is the proper term. It takes input and applies it intelligently to future tasks. To me "learning" is loading up the memory. "Thinking" is more like applying it intelligently, which is not exactly the same, plus it's a subsequent phase. Or at least a dependent one with latency. >Machines can learn, machine learning has been around for decades. Learning doesn’t require biology. Now all this sounds really straightforward. >Machines don't learn. They encode, compress and record. I can agree with this too, people are lucky they have more than kilobytes to work with or they'd be compressing like there's no tomorrow. But regardless, eventually the memory fills up or the data runs out and then you have to do something with it, whether very intelligent or not. Might as well anthropomorphize my dang self. If you know who Kelly Bundy is, she's a sitcom character representing a young student of limited intellect and academic interest. Part of the schtick was that she verbally reported the error message when her "brain is full" ;) It was just an observation, no thinking was required or implied ;) If the closest a machine is going to come is when its memory is filled, so be it. What more can you expect anyway from a mere machine during the fundamental data input process? If that's the nearest you're going to get to organic learning, that'll have to serve as "machine learning" until more sensible nuance comes along. Memory can surely be filled more intelligently sometimes than others, which should make a huge difference in how intelligently the data can handled afterward, plus some data is bound to be dramatically more useful than others too. But the real intelligent stuff is supposed to be the processing done with this data in an open-ended way after all those finite elements have been stored electronically. To me the "learning" is the data input, or "student" phase, and the intelligence is what you do with those "learnings" if it can be made smart. It can be good to build differing scenarios from the exact same data, and ideally become better at decision-making through time without having more raw data come in. Improvements like this would be the next level of learning so now you've got more than just the initial data-filling. As long as there's room in the memory, otherwise you're going to have to "forget" something first when your brain is already full :) I just don't think things are ideal myself. >journal on extra-terrestrial aliens would prove that the word "aliens" is used to mean "extra-terrestrials" Exactly, it proves that the terminology exists, and gives it more meaning sometimes. The aliens don't have to be as real as you would like, or even exist, nor the intelligence.

tsimionescu 2y ago

The law on copyright doesn't depend on the word "learning", it depends on whether it's a human doing it, or a mechanical process. If a human reads a book and produces a different book that's sort-of-derivative but doesn't copy too many elements too directly, then that book is a new creative work and doesn't infringe on the copyright of the original author. For example, 50 Shades of Gray is somewhat derivative of Twilight (famously starting as a Twilight fan-fic) but it's legally a separate copyright. Conversely, if you use a machine to produce the same book, taking only copyrighted text as input and an algorithm that replaces certain words and phrases and adds certain passages, then the result is a derivative work of the original and it infringes the copyright of the original author. So again, the facts of the law are pretty simple, at the moment at least: even if a machine and a human do the exact same thing, it's still different from a legal perspective.

DoctorOetker 2y ago
machines work on behalf of humans, any human could similarly claim their typewriter did it
- tsimionescu 2y ago
  That's still irrelevant. The distinction the law makes is between human creativity vs machine processes. A human may or may not have added their own creativity to the mix, and the law falls on the side of assuming creativity from humans (though if there is proof that the human followed some mechanical process, that likely changes things). Machines are not creative, by definition in the law today, so any transformation that a machine makes is irrelevant, the work is fundamentally the same work that was used as input from a legal perspective. That this work is done on behalf of a human changes nothing, the problem the law has with this is that the human is copying the original work without copyright, even if the human used a machine to produce an altered copy. Whether they used bcrypt, zip, an mp3 encoder, a bad copy machine, or a machine learning algorithm, the result is the same: the output of a purely mechanical process is still a copy of the original works from a copyright perspective.