Re: GCC GSoC 2026: Call for project ideas and mentors
Ville Voutilainen <[email protected]>
| Newsgroups | gmane.comp.gcc.rust,gmane.comp.gcc.fortran,gmane.comp.gcc.devel,gmane.comp.gcc.libstdc++.devel |
|---|---|
| Message-ID | <CAFk2RUZ_Y5t=cUaNvdyq_GdHKnWd0nD9HVYskvAK835h-MO5zw@mail.gmail.com> |
On Tue, 20 Jan 2026 at 15:12, Richard Kenner <[email protected]> wrote: > > > "It is not permitted to use Large Language Model (such as ChatGPT, > > Gemini, Copilot etc.) to generate any non-trivial (legally > > significant) part of code that is produced as a part of GCC GSoC." > > Actually, let me get back to this because there's an ambiguity here. > We certainly have no issue with people using an LLM to assist in their > work (by answering questions). "Used to generate" could include this > scenario. The issue isn't whether they are *using* an LLM, but about > including *code generated* by an LLM. (And I had a typo.) So let me > change my suggestion to: > > With the exception of test cases, you may not include any non-trivial > (more than de-minimus) amount of code generated by a Large Language > Model (an "AI" such as ChatGPT, Gemini, and Copilot.) in code that you > submit as a participant of GCC GSoC. FWIW, I don't find the original suggestion ambiguous. "Use to generate" is very different from "Use to learn how to write".