4 ms·

I feel this argument is weakened by the fact that no one training these models feeds their own proprietary source code into publicly available models. Practica

I feel this argument is weakened by the fact that no one training these models feeds their own proprietary source code into publicly available models. Practically speaking I feel this is a strong hint that some degree of copyright laundering is taking place.

ShamelessC 3y ago
> no one training these models feeds their own proprietary source code into publicly available models You are presupposing that the company’s own code would somehow be a massive boon for the model, resulting in lower loss overall. In reality, it would skew the model towards that company’s “mode” of coding which isn’t what “normal” programmers expect. In fact, they are most likely to expect the coding styles they learned from, and that is most likely to be found in public examples (GitHub, StackOverflow, textbooks, Reddit, etc.) This argument is so silly to me. Anyone who has worked at a large enterprise, whether it’s Google, Amazon or Target, knows that company code is effectively guaranteed to be extremely hard to work with. This happens for organizational reasons and really the best thing to do about it is admit that it’s happening rather than pretend it’s all perfect.
- brvsft 3y ago
  The reason they don't train the model on their code is specifically because they don't want it accidentally spitting out snippets of their proprietary code, not because the code is "extremely hard to work with." I'm amazed you called that argument silly while countering with this.
  - ShamelessC 3y ago
    It is because of both and I agree that your reasoning takes clear precedence. I was merely pointing out the good faith position that “even if they wanted to, they wouldn’t do it”. That definitely wasn’t very clear from my comment however.
smallnix 3y ago
> CodeWhisperer is a generative AI service powered by a foundation model trained on various data sources, including Amazon and open-source code.
Gigachad 3y ago
I don’t think that suggests copyright infringement, more that you could do something like “request from the Amazon internal admin page url” and the AI will generate you the correct link. Or tell you other info which is private. The AI hasn’t copied anything, but it does know secrets and can use them. This does not matter for open source since anyone can see this info already.
circuit10 3y ago
You also wouldn't see them giving their code to people learning to code to read and learn from, but that doesn't necessarily mean that learning from something violates copyright. It seems like a bit of a grey area where the distinction is between learning and using it directly
- lelanthran 3y ago
  > You also wouldn't see them giving their code to people learning to code to read and learn from, Actually, we do see them do exactly that. Microsoft is ahppy to have a shared-source licence that gives away much of the Windows source code to universities. The fact that they don't want to train their models on it says a lot. If, as they claim, learning from existing materials does nto devalue those materials in any way, they'd chuck the entirety of the Windows source code at the LLM for training purposes.

ShamelessC 3y ago

> no one training these models feeds their own proprietary source code into publicly available models You are presupposing that the company’s own code would somehow be a massive boon for the model, resulting in lower loss overall. In reality, it would skew the model towards that company’s “mode” of coding which isn’t what “normal” programmers expect. In fact, they are most likely to expect the coding styles they learned from, and that is most likely to be found in public examples (GitHub, StackOverflow, textbooks, Reddit, etc.) This argument is so silly to me. Anyone who has worked at a large enterprise, whether it’s Google, Amazon or Target, knows that company code is effectively guaranteed to be extremely hard to work with. This happens for organizational reasons and really the best thing to do about it is admit that it’s happening rather than pretend it’s all perfect.

brvsft 3y ago
The reason they don't train the model on their code is specifically because they don't want it accidentally spitting out snippets of their proprietary code, not because the code is "extremely hard to work with." I'm amazed you called that argument silly while countering with this.
- ShamelessC 3y ago
  It is because of both and I agree that your reasoning takes clear precedence. I was merely pointing out the good faith position that “even if they wanted to, they wouldn’t do it”. That definitely wasn’t very clear from my comment however.

smallnix 3y ago

> CodeWhisperer is a generative AI service powered by a foundation model trained on various data sources, including Amazon and open-source code.

Gigachad 3y ago

I don’t think that suggests copyright infringement, more that you could do something like “request from the Amazon internal admin page url” and the AI will generate you the correct link. Or tell you other info which is private. The AI hasn’t copied anything, but it does know secrets and can use them. This does not matter for open source since anyone can see this info already.

circuit10 3y ago

You also wouldn't see them giving their code to people learning to code to read and learn from, but that doesn't necessarily mean that learning from something violates copyright. It seems like a bit of a grey area where the distinction is between learning and using it directly

lelanthran 3y ago
> You also wouldn't see them giving their code to people learning to code to read and learn from, Actually, we do see them do exactly that. Microsoft is ahppy to have a shared-source licence that gives away much of the Windows source code to universities. The fact that they don't want to train their models on it says a lot. If, as they claim, learning from existing materials does nto devalue those materials in any way, they'd chuck the entirety of the Windows source code at the LLM for training purposes.