[gdal-dev] Call for review: RFC111: AI/LLM tool policy

[Even Rouault]

Le 09/02/2026 à 21:57, Kurt Schwehr a écrit :
> It looks pretty good to me.
>
> One comment:
>
> What about an AI that autonomously finds bugs that trigger sanitizers 
> that are verifiable like OSS Fuzz does?

I'd say that's generally fine, because at the end you have a "digital" 
human (the sanitizer), that is something with inductive logic, checking 
the output of the AI. But there must also be some protection to avoid 
such tools to spam our issue tracker with hundreds of reports, ie it 
should first prioritize the most critical ones and make sure they are 
addressed before issuing reports of lesser criticity.  It is hard to 
encode the rules besides trying to apply empathy: try to put yourself at 
the receiving end of the flow and imagine how you'd react


-- 
http://www.spatialys.com
My software is free, but my time generally not.