Date: Mon, 13 Oct 2025 21:43:14 +0200 From: Steffen Nurpmeso <steffen@sdaoden.eu> To: Michael Butler <imb@protected-networks.net> Cc: stable@freebsd.org Subject: Re: What to do, if chatgpt gives a wrong answer Message-ID: <20251013194314.eQKOstcJ@steffen%sdaoden.eu> In-Reply-To: <d716c704-3ae7-4c6a-a37b-fcecc3b743cf@protected-networks.net> References: <2ee60f98af9e62da887deb970006a86dfe19a207.camel@ft-c.de> <865xci7udf.fsf@ltc.des.dev> <d716c704-3ae7-4c6a-a37b-fcecc3b743cf@protected-networks.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Michael Butler wrote in <d716c704-3ae7-4c6a-a37b-fcecc3b743cf@protected-networks.net>: |On 10/13/25 10:23, Dag-Erling Sm=C3=B8rgrav wrote: |> ft <ml@ft-c.de> writes: |>> When I ask ChatGPT a question [...] I very often get the wrong answer. |>=20 |> Don't do that, then. |>=20 |> Artificial intelligence exists only in science fiction. ChatGPT is not |> artificial intelligence, it is a deterministic computer program that |> uses a large statistical model of natural language to complete text |> submitted to it. Since it has been trained on a corpus that includes |> many examples of questions followed by answers, when prompted with a |> question, it will produce something that has the shape of an answer, but |> absolutely no effort has gone into ensuring that the answer is correct, |> nor does anyone involved have any idea how to even begin doing that. | |Spotted this gem on another mailing list recently .. | |All that *ANY* LLM can provide here is a statistically=20 |less-improbable-than-random stream of words that may or may not include= =20 |any statements of true facts, uninfluenced by anything resembling knowle\ |dge. Now i have to say one thing. I never tried AI, i do not like AI, because it is used and trained by the same old dumb humans, not to talk about Kubrick's HAL, and of course Asimov, i *think* AI should be a scientific thing that is carefully developed before it enters "the normal world", maybe even so that dedicated wind and solar farms are built in order to drive the AI then used in "the normal world". Having said that. I recently opened a ChatGPT instance to read the conversation initiated by the Field medalist and otherwise Hyper Mathematic whose name i have forgotten (not a mathematician here) who tried it out in order to address a problem asked by someone on some "stackoverflow-alike-thing for mathematicians". It was about proof that, iirc, "the sum of dividers of a number is always larger than the number itself". Now i tell you, that shitty conversion was fascinating, not that i understood mostly a single mathematical term they were throwing back and forth, and i would not post this message if then, and here i was stunned and still, one answer that cames back was like about "love this topic" or similar. And then it crushed to super detail, and in the end it presented a small python reproducer. In the hands of, and in correspondence with this math genius the AI worked in an amazing way, with turns that have shown a thrilling -- as i with my own restricted capabilities think -- topic reflection. This does not counteract the first paragraph. --steffen | |Der Kragenbaer, The moon bear, |der holt sich munter he cheerfully and one by one |einen nach dem anderen runter wa.ks himself off |(By Robert Gernhardt)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20251013194314.eQKOstcJ>