Another example of how AI is bad, bug fix for 20875

[Indraj Gandham]

Apologies, I should have been more clear. I meant to refer only to the use of "Co-Authored-By". I did not intend to suggest that it is acceptable not to disclose LLM use.

However, I would point out that this field alone does not necessarily provide enough information to know how the author used the LLM or what proportion of a submission was generated by an LLM.

I also don't think that it's a given that disclosing the use of LLMs would dilute any copyright risks; on the contrary I think it might make it worse. It has been shown that LLMs can reproduce copyrighted works and I don't believe that a court would invalidate any copyright claim arising from the use of such output. Even reproducing the structure and other "non-literal" elements of copyrighted programs could potentially result in penalties. This was established during the course of Computer Associates Inc. v Altai Inc. in 1992 even though it was determined that no infringement had occurred in that specific case.

To what extent this is enforceable for LLM output is up for debate, since neither the user nor the copyright holder may be aware of the infringement. Nevertheless, relying on a blanket exemption incurs some risk, even if it is difficult to quantify.