On Sun, Mar 29, 2026 at 12:15:33PM +0300, Jean Louis wrote: > * [email protected] <[email protected]> [2026-03-28 14:39]: > > On Sat, Mar 28, 2026 at 02:08:31PM +0300, Jean Louis wrote: > > > > [...] > > > > > Yeah, you're not wrong—LLMs will absolutely bullshit eloquently if you > > > let them. The literate docs don't magically fix that. > > > > [...] > > > > The (AFAICS) unsolved problem is that there is no way to be > > sure that the (eloquent) text corresponds to the code. If > > not, it would be highly counterproductive. > > How does the uncertainty of machine-generated code compare to the > uncertainty of human-written code? > > I suggest you try it.
[...] I have been debugging other people's code for a while now, and yes, comments and docs aren't always right (they tend to age, and since the test suite doesn't reach them...). Thing is, after a while in the code I tend to have a mental model (call it "wet LLM" if you fancy) of the typically small group of persons who were at the code and of the processes which shaped it. I doubt I'll be able to do the same with LLM generated code. I predict: LLM generated code will be more "throwaway". Cheers -- t
signature.asc
Description: PGP signature
--- via emacs-tangents mailing list (https://lists.gnu.org/mailman/listinfo/emacs-tangents)
