Re: Coding assistants (LLM) policy

(Resending because I didn't send it to the list, oops)

There is an use of LLM I haven't seen mentioned in the discussion and is 
different from the 'ask LLM to generate code' case. The policy seems to 
implicitly assume 'generate code/...'.

Some time ago, I read something on lwn about asking a specialised LLM to 
search for bugs and going by the report it actually did decently.

It's not with its own potential problems (e.g. blindly assuming the 
model's interpretation is correct, keep arguing until it gives in, 
assuming 100% true-positive and true-negative rate, flooding bug tracker 
with duplicate reports ...), but those are largely problems shared by 
things like linters.

Many of the reasons to discourage 'generative LLM' don't seem to apply 
to 'LLM as fancy broad-spectrum linter'.

I do not know whether or not 'extreme electricity use' (environmental 
impact, ...) applies to this other use case.

Best regards,
Amélie Devos