5 ms·

Google Announces Stealing Part of a Production Language Model

117 points by ludovicianul 3y ago

tellarin 3y ago
Discussion: https://news.ycombinator.com/item?id=39675735 https://news.ycombinator.com/item?id=39675735
gmerc 3y ago
I like how they use stealing .... while maintaining steadfast that it's fair use when they use other people's data to train the model.
- baryphonic 3y ago
  Ironically, weights and biases are likely not copyrightable.[0] I looked into this first when the LLaMA model was released on BitTorrent, and concluded that downloading it was perfectly legal. The courts even have controlling precedents that say "sweat of the brow" is not a valid justification for copyright, meaning that the effort these large organizations put into training their models does not grant them protection.[1] [0]https://news.ycombinator.com/item?id=35121093 https://news.ycombinator.com/item?id=35121093 [1]https://definitions.uslegal.com/s/sweat-of-the-brow-doctrine/ https://definitions.uslegal.com/s/sweat-of-the-brow-doctrine...
  - fl7305 3y ago
    > Ironically, weights and biases are likely not copyrightable. > The courts even have controlling precedents that say "sweat of the brow" is not a valid justification for copyright You might be right for the US, but the EU seems to have different laws: https://europa.eu/youreurope/business/running-business/intellectual-property/database-protection/index_en.htm https://europa.eu/youreurope/business/running-business/intel...
    - baryphonic 3y ago
      That's fair enough, but Meta, Google, OpenAI, etc. are all American companies, thus US law would likely apply.
      - azinman2 3y ago
        Depends on which part of the world you’re in.
      - dymk 3y ago
        Not when they operate in other countries
  - lostmsu 3y ago
    IANAL, but I would not immediately agree with the first link, because I believe their reasoning applies to any video encoded with loss.
- HarHarVeryFunny 3y ago
  Not sure what their motivation was to use the word "stealing" here - they are obtaining the output layer weights (mapping from hidden dimension to output), but it seems the hidden dimension size is the most interesting thing here. They could have called it probing rather than stealing.
  - HumblyTossed 3y ago
    > Not sure what their motivation was to use the word "stealing" here Maybe, simply, to get peoples' attention?
- bodiekane 3y ago
  Which maps perfectly onto how people would talk about similar work if it wasn't AI. If I go to the library and read a bunch of books and journal articles to understand something, that's certainly research or education and not "stealing". If I then use my understanding gained from that education or training to create a new invention, it's plausible that someone could "steal" information from my invention that I never intended to publish, and which was reasonably expected to be private trade secret information not available to others.
  - LtWorf 3y ago
    The moment that you can train an AI to be a good developer just with a java manual and some slides on design patterns, your analogy will make sense. But for now an AI requires billion of times more data than a human brain to extrapolate anything useful. Which means that your comparison is meaningless.
    - GlickWick 3y ago
      I think it's a meaningful comparison still. As a human I can comb through tons of prior art to learn my trade, and that's how we learn most of what we know. Yes, an AI requires more, but it's still the same basic idea. I'm still not comfortable with using AI to replace artistic functions in particular for a variety of reasons, but the person you're replying to is definitely within the realm of reason.
      - LtWorf 3y ago
        > it's still the same basic idea You create weights to decide which word comes next most probably inside your brain and need to do million of multiplications to say a word? I think it might not be the same basic idea at all.
        GlickWick 3y ago
        I think you might want to take a step back and understand the abstraction they were making.
cuu508 3y ago
Better link: https://arxiv.org/abs/2403.06634 https://arxiv.org/abs/2403.06634
nippoo 3y ago
I'm so curious what an example set of queries actually looks like here. Like - what were the actual prompts used!
fragmede 3y ago
[dupe], discussed previously at https://news.ycombinator.com/item?id=39675735 https://news.ycombinator.com/item?id=39675735 The twitter post adds no new information and poorly links to the same paper.
hakdbha 3y ago
[dead]

117 points by ludovicianul 3y ago

Discussion: https://news.ycombinator.com/item?id=39675735 https://news.ycombinator.com/item?id=39675735

I like how they use stealing .... while maintaining steadfast that it's fair use when they use other people's data to train the model.

baryphonic 3y ago
Ironically, weights and biases are likely not copyrightable.[0] I looked into this first when the LLaMA model was released on BitTorrent, and concluded that downloading it was perfectly legal. The courts even have controlling precedents that say "sweat of the brow" is not a valid justification for copyright, meaning that the effort these large organizations put into training their models does not grant them protection.[1] [0]https://news.ycombinator.com/item?id=35121093 https://news.ycombinator.com/item?id=35121093 [1]https://definitions.uslegal.com/s/sweat-of-the-brow-doctrine/ https://definitions.uslegal.com/s/sweat-of-the-brow-doctrine...
- fl7305 3y ago
  > Ironically, weights and biases are likely not copyrightable. > The courts even have controlling precedents that say "sweat of the brow" is not a valid justification for copyright You might be right for the US, but the EU seems to have different laws: https://europa.eu/youreurope/business/running-business/intellectual-property/database-protection/index_en.htm https://europa.eu/youreurope/business/running-business/intel...
  - baryphonic 3y ago
    That's fair enough, but Meta, Google, OpenAI, etc. are all American companies, thus US law would likely apply.
    - azinman2 3y ago
      Depends on which part of the world you’re in.
    - dymk 3y ago
      Not when they operate in other countries
- lostmsu 3y ago
  IANAL, but I would not immediately agree with the first link, because I believe their reasoning applies to any video encoded with loss.
HarHarVeryFunny 3y ago
Not sure what their motivation was to use the word "stealing" here - they are obtaining the output layer weights (mapping from hidden dimension to output), but it seems the hidden dimension size is the most interesting thing here. They could have called it probing rather than stealing.
- HumblyTossed 3y ago
  > Not sure what their motivation was to use the word "stealing" here Maybe, simply, to get peoples' attention?
bodiekane 3y ago
Which maps perfectly onto how people would talk about similar work if it wasn't AI. If I go to the library and read a bunch of books and journal articles to understand something, that's certainly research or education and not "stealing". If I then use my understanding gained from that education or training to create a new invention, it's plausible that someone could "steal" information from my invention that I never intended to publish, and which was reasonably expected to be private trade secret information not available to others.
- LtWorf 3y ago
  The moment that you can train an AI to be a good developer just with a java manual and some slides on design patterns, your analogy will make sense. But for now an AI requires billion of times more data than a human brain to extrapolate anything useful. Which means that your comparison is meaningless.
  - GlickWick 3y ago
    I think it's a meaningful comparison still. As a human I can comb through tons of prior art to learn my trade, and that's how we learn most of what we know. Yes, an AI requires more, but it's still the same basic idea. I'm still not comfortable with using AI to replace artistic functions in particular for a variety of reasons, but the person you're replying to is definitely within the realm of reason.
    - LtWorf 3y ago
      > it's still the same basic idea You create weights to decide which word comes next most probably inside your brain and need to do million of multiplications to say a word? I think it might not be the same basic idea at all.
      - GlickWick 3y ago
        I think you might want to take a step back and understand the abstraction they were making.

cuu508 3y ago

Better link: https://arxiv.org/abs/2403.06634 https://arxiv.org/abs/2403.06634

nippoo 3y ago

I'm so curious what an example set of queries actually looks like here. Like - what were the actual prompts used!

fragmede 3y ago

[dupe], discussed previously at https://news.ycombinator.com/item?id=39675735 https://news.ycombinator.com/item?id=39675735 The twitter post adds no new information and poorly links to the same paper.

hakdbha 3y ago

[dead]