On Sat, 21 Mar 2026 14:43:19 +0100,
Thomas de Grivel <thodg@kmx.io> wrote:
> 
> Maybe intersting points to remember : if you have theoretical
> knowledge of kernel-land an AI has good insights on the workings of
> the C code.
> 

LLM are the perfect plagiarism machines.

If you have narrow enough knowledge's domain, it will likley reproduce the
only known things to it.

It never "thinks" or "reasons", it try to predict the next token.

If the next tokens are nonsense and noise, then it won't work as the code,
but if it works, it is probably copy and paste from someone else's code.

Here some example how folks extracted an actual Harry Potter text:
https://arxiv.org/abs/2601.02671

-- 
wbr, Kirill