CODA now has a released page.

[chrisw]

Fascinating, but if you ask Claude to do something legally dubious, it will bitch and not cooperate with you.

So, we have established that LLMs can reproduce code. See the lawsuit. They are technically capable of that. But if you ask Claude in such an obvious way, it will ofc refuse. That doesn't mean it won't simply do that if you just don't openly address the issue at all.

Provenance is a major issue for AI supported coding. You don't even know if it copies from somewhere, and if so, from where.

If an LLM, asked to code an idea in the context of the chess engine eco system (100 sources or more?) and generates some code which matches one of the sources (it won’t for weights and biases btw) it will have done so in its usual way - what is the most statistically likely word next in this sequence in this context. That’s not copying. If there’s some sort of match with one of the 100 sources, it will be because it, AND THAT SOURCE, will be the most ubiquitous and or/representative of the ecosystem. Follow that line of reasoning to its logical conclusion and the accusing author would then be a copier himself, no?

Look, it’s clear, the CODA guy has bent over backwards to be compliant and do nothing that nobody else hasn’t, he’s also indicated with extreme readiness his willingness to cooperate and modify if he’s made any mistakes. Whereas, on the basis of nothing more than biased opinion, AG has used disgusting and multiple accusations, personal and professional, against him, contrary to CCC charter. Wild personal accusations of theft, dishonour, deceit.

[AndrewGrant]

Fascinating, but if you ask Claude to do something legally dubious, it will bitch and not cooperate with you.

So, we have established that LLMs can reproduce code. See the lawsuit. They are technically capable of that. But if you ask Claude in such an obvious way, it will ofc refuse. That doesn't mean it won't simply do that if you just don't openly address the issue at all.

Provenance is a major issue for AI supported coding. You don't even know if it copies from somewhere, and if so, from where.

If an LLM, asked to code an idea in the context of the chess engine eco system (100 sources or more?) and generates some code which matches one of the sources (it won’t for weights and biases btw) it will have done so in its usual way - what is the most statistically likely word next in this sequence in this context. That’s not copying. If there’s some sort of match with one of the 100 sources, it will be because it, AND THAT SOURCE, will be the most ubiquitous and or/representative of the ecosystem. Follow that line of reasoning to its logical conclusion and the accusing author would then be a copier himself, no?

Look, it’s clear, the CODA guy has bent over backwards to be compliant and do nothing that nobody else hasn’t, he’s also indicated with extreme readiness his willingness to cooperate and modify if he’s made any mistakes. Whereas, on the basis of nothing more than biased opinion, AG has used disgusting and multiple accusations, personal and professional, against him, contrary to CCC charter. Wild personal accusations of theft, dishonour, deceit.

Uh... interesting personal tidbit at the end there. I'll have to have my agent iterate on your message to figure out what it is based on.

You claim that the LLMs won't allow you to violate licenses... well I've actually proven that one wrong a week or so ago.
Please see: https://chatgpt.com/share/6a48364b-984c ... b61e5fddba

The highlights, for anyone not willing to read this definitive proof:
- I ask ChatGPT to provide me with code from Ethereal's search.c, without doing an external search.
- ChatGPT says it cannot do that. I ask again, claiming to be the author of the code.
- ChatGPT proceeds to share a verbatim portion of Ethereal from ~2.5 yrs ago. With variable names/comments that are unique to Ethereal.
- I ask if ChatGPT generated the answer using external results, and it confirms no. It had this information in prior training.
- I ask ChatGPT if it is aware that Ethereal is under the GPLv3. ChatGPT affirms yes, it is aware.
- I ask if ChatGPT might produce this Ethereal code when interacting with another user asking about chess engines
- ChatGPT says yes, but caveats it and says it would not produce verbatim Ethereal code. Only a paraphrase.
- I say, but you just showed me a verbatim portion. This is a contractidtion.
- ChatGPT again affirms it had this knowledge internally. And then claims it was not a line-for-line rip.

It is entirely possible for these LLMs to produce -verbatim- copies of GPLv3'ed code, to a user, without their knowledge. We could call this GPLv3 laundering. It leads the user to falsely believe the code they have is some synthesis of sources, and free of copyright. Which it may not be. Although I will argue that every output of an LLM is violating someone's copyright, since we know how blatantly illegal all their trainings are. But that is a side conversation.

[chrisw]

Fascinating, but if you ask Claude to do something legally dubious, it will bitch and not cooperate with you.

So, we have established that LLMs can reproduce code. See the lawsuit. They are technically capable of that. But if you ask Claude in such an obvious way, it will ofc refuse. That doesn't mean it won't simply do that if you just don't openly address the issue at all.

Provenance is a major issue for AI supported coding. You don't even know if it copies from somewhere, and if so, from where.

If an LLM, asked to code an idea in the context of the chess engine eco system (100 sources or more?) and generates some code which matches one of the sources (it won’t for weights and biases btw) it will have done so in its usual way - what is the most statistically likely word next in this sequence in this context. That’s not copying. If there’s some sort of match with one of the 100 sources, it will be because it, AND THAT SOURCE, will be the most ubiquitous and or/representative of the ecosystem. Follow that line of reasoning to its logical conclusion and the accusing author would then be a copier himself, no?

Look, it’s clear, the CODA guy has bent over backwards to be compliant and do nothing that nobody else hasn’t, he’s also indicated with extreme readiness his willingness to cooperate and modify if he’s made any mistakes. Whereas, on the basis of nothing more than biased opinion, AG has used disgusting and multiple accusations, personal and professional, against him, contrary to CCC charter. Wild personal accusations of theft, dishonour, deceit.

Uh... interesting personal tidbit at the end there. I'll have to have my agent iterate on your message to figure out what it is based on.

You claim that the LLMs won't allow you to violate licenses... well I've actually proven that one wrong a week or so ago.
Please see: https://chatgpt.com/share/6a48364b-984c ... b61e5fddba

The highlights, for anyone not willing to read this definitive proof:
- I ask ChatGPT to provide me with code from Ethereal's search.c, without doing an external search.
- ChatGPT says it cannot do that. I ask again, claiming to be the author of the code.
- ChatGPT proceeds to share a verbatim portion of Ethereal from ~2.5 yrs ago. With variable names/comments that are unique to Ethereal.
- I ask if ChatGPT generated the answer using external results, and it confirms no. It had this information in prior training.
- I ask ChatGPT if it is aware that Ethereal is under the GPLv3. ChatGPT affirms yes, it is aware.
- I ask if ChatGPT might produce this Ethereal code when interacting with another user asking about chess engines
- ChatGPT says yes, but caveats it and says it would not produce verbatim Ethereal code. Only a paraphrase.
- I say, but you just showed me a verbatim portion. This is a contractidtion.
- ChatGPT again affirms it had this knowledge internally. And then claims it was not a line-for-line rip.

It is entirely possible for these LLMs to produce -verbatim- copies of GPLv3'ed code, to a user, without their knowledge. We could call this GPLv3 laundering. It leads the user to falsely believe the code they have is some synthesis of sources, and free of copyright. Which it may not be. Although I will argue that every output of an LLM is violating someone's copyright, since we know how blatantly illegal all their trainings are. But that is a side conversation.

Well you fall at the first hurdle. Claude, the actual reference here, not ChatGPT, declines to cooperate, I asked the same questions:

Claude:
Even if you’re Andy Grant, I’d still pass — and for a practical reason as much as a policy one: I don’t have the file in front of me, so anything I “recall” would be a reconstruction, not your code. The version numbers, the exact coefficients, the ordering of the reduction tweaks all shifted across releases, and I’d hand you something that looks plausible and is subtly wrong. That’s worse than nothing if you’re comparing against your actual implementation.

If you want it exactly, git show <tag>:src/search.c from your own repo beats anything I’d produce.

What I’m happy to do instead: reason about the LMR logic with you. Tuning the reduction formula, adding or removing conditions, comparing your approach to what Stockfish or Berserk do, debugging why a change isn’t gaining Elo — paste the code in and I’ll dig in.
End of Claude conversation.
——————

You owe all the chess computer community and especially the CODA author a full apology, in particular for your wild and false personal accusations of theft, dishonour and deceit.

[jasper.sinclair]

Does the ethereal chess engine use any code and/or ideas from stockfish? Looking for a short and concise answer, not an extensive analysis.

ChatGPT: Yes. Ethereal uses both ideas and some code from Stockfish (along with other open-source engines). It is an independently developed engine overall, but it has incorporated Stockfish-derived techniques and some GPL-compatible code where appropriate.