On Sun, Mar 29, 2026 at 01:58:54PM +0300, Jean Louis wrote: > * Dr. Arne Babenhauserheide <[email protected]> [2026-03-29 12:58]: > > > True certainty can be obtained by testing functions and seeing if they > > > are doing what is meant to be. > > > > 1. That’s only true if the tests are completely exhaustive. Which very > > likely makes them a lot longer than the code itself. > > Same for human code. > > > 2. Who reviews that exhaustive test code? > > I am used to making functions small and making sure they work. Each > programmer deals with this in their own way. But the benefits of LLM > generation are immense. > > Planning (as generated by the LLM) alone is of immense use — it can > surface things you weren't even thinking about. You don't have to > implement the plan. Just look at it. If it's useful, use it. If not, > discard it. > > The benefits of LLM generation are undisputed. We're just talking into > the air here. Thousands and thousands of people are building new > applications every day. What isn't suitable for your specific case? > Just discard it.
Don't get me wrong. There are places I see the use for stochastically driven, model-informed generators. Testing is one -- you have a space too big to cover exhaustively (-> sqlite testing :) so your only choice is to poke at it, by informed guesses or by throwing dice. The second might help you, if you're lucky (heh) to stumble upon stuff your guess wasn't as informed as you thought in the first place. Those things have existed long before "AI" was a thing. Monte Carlo simulations. Fuzzers. You name them. Very useful. The innovation with "AI" is that the "model-informed" part now comes from a "big data" model, built on passing the vacuum cleaner over "The Internet". Fine. Still useful. But use with care. I don't want it to enter free sofware development (which is, as software development always is, but even more, a higly social endeavour) without big, fat guardrails. Especially not these days, where companies desperate to save investor money want to make us believe (and are partially succeeding!) that the thing is, in a way, somewhat human. This is poison for the community. Cheers -- t
signature.asc
Description: PGP signature
--- via emacs-tangents mailing list (https://lists.gnu.org/mailman/listinfo/emacs-tangents)
