RFC: The Binutils and LLM generated patches

[Richard Earnshaw (lists)]

On 06/10/2025 10:59, Siddhesh Poyarekar wrote:
> On 2025-10-06 05:28, Richard Earnshaw (lists) wrote:
>>>> I think using an LLM to help improve the /readability/ of documentation should also be OK.  The key point though is that the primary copyrightable step comes from the initial authoring and that should be done by the human.
>>>
>>> Yes, but I think that tools for fixing layout, finding spelling errors, etc., are not really AI and that sort of jobs is not one for LLMs at all.  Those are fixed function well defined tasks done with traditional programs that are obvious state transformers, where IP issues cannot arise.
>>
>> I'm not talking about traditional tools (like ispell), or even the minor grammar suggestions that tools like Office have had for years, but more of AI tools that can take your text and write a completely new version that has better structure and clarity.  I've seen tools that can do that and they can, when used with care, make a huge difference to the readability of a document.
>>
> 
> Do we need the FSF to step in and assure us that restructuring of text (as long as the meaning is preserved) is not a legally significant change?
> 
> Thanks,
> Sid

If you're going to ask a lawyer a question, be sure you're asking the /right/ question :)

In this case it's not so much whether the change is copyrightable (copyright is about expression of an idea after all), as whether the generated text contains a meaningful (potentially identifiable) use of other copyrighted material.

Given the enormous corpus of English text that AI's consume, I very much doubt that would be the case, but it is fair to ask if the FSF would have concerns in that area.

R.