<![CDATA[This is so relevant.]]>RE: https://radikal.social/@malte/117354154268154895

This is so relevant.

But the lesson is that AI is not 'worse' giving answers in a field he has knowledge in. It's just that in that field, he can be absolutely sure it IS wrong.

]]>https://forum.fedi.dk/topic/51f26905-d6f8-47f5-9757-ad69d710800c/this-is-so-relevant.RSS for NodeThu, 01 Oct 2026 12:31:54 GMTTue, 29 Sep 2026 12:23:06 GMT60<![CDATA[Reply to This is so relevant. on Tue, 29 Sep 2026 20:48:12 GMT]]>@Remittancegirl Yes, and I'll add that for me this tells me that the only fields I would want to actually use any generative LLM - if at all - are the fields where I will be able to tell with certainty whether the answer is right or wrong. I learned a lot from taking in @pluralistic's reflections on the concept of 'discernment' in this case - it's still the most integrative take on this problem I've read: https://pluralistic.net/2026/07/28/hitl-ers/

]]>https://forum.fedi.dk/post/https://radikal.social/users/malte/statuses/117356291190471903https://forum.fedi.dk/post/https://radikal.social/users/malte/statuses/117356291190471903Tue, 29 Sep 2026 20:48:12 GMT<![CDATA[Reply to This is so relevant. on Tue, 29 Sep 2026 14:38:27 GMT]]>@Remittancegirl Gell-Mann amnesia for the digital era.

]]>https://forum.fedi.dk/post/https://mastodon.social/users/bransonturner/statuses/117354837269983775https://forum.fedi.dk/post/https://mastodon.social/users/bransonturner/statuses/117354837269983775Tue, 29 Sep 2026 14:38:27 GMT<![CDATA[Reply to This is so relevant. on Tue, 29 Sep 2026 13:53:29 GMT]]>@Remittancegirl Exactly. We don't know everything, but if this 'hallowed oracle of knowledge' has shown it's at that same level, it's hard to know what it knows, so I'll default it to 'you're wrong'. Because these things will also tell you the biggest lie and they'll try to sell that as the truth, if you just mess enough with those systems.

I'll take the long way home and look things up myself, even if I have to go to a library and go through books.

]]>https://forum.fedi.dk/post/https://writing.exchange/users/paulk/statuses/117354660443966696https://forum.fedi.dk/post/https://writing.exchange/users/paulk/statuses/117354660443966696Tue, 29 Sep 2026 13:53:29 GMT<![CDATA[Reply to This is so relevant. on Tue, 29 Sep 2026 13:45:08 GMT]]>@paulk And the moment it displays itself as untrustworthy, it is useless. And not just for one subject, but for all of them. Not because it's always wrong, but because it could be wrong about anything.

]]>https://forum.fedi.dk/post/https://mstdn.social/users/Remittancegirl/statuses/117354627632907066https://forum.fedi.dk/post/https://mstdn.social/users/Remittancegirl/statuses/117354627632907066Tue, 29 Sep 2026 13:45:08 GMT<![CDATA[Reply to This is so relevant. on Tue, 29 Sep 2026 13:31:55 GMT]]>@Remittancegirl I don't understand why people don't test this straight away, before using it for anything else. The first thing I did was ask it a few questions like "what books has timtfj written?" (none, but it gave me a list of interesting looking ones) and enjoy the resulting nonsense. It was a language experiment, like the line-and-a-word game we had in our family when I was a child.

]]>https://forum.fedi.dk/post/https://mastodon.social/users/timtfj/statuses/117354575682176985https://forum.fedi.dk/post/https://mastodon.social/users/timtfj/statuses/117354575682176985Tue, 29 Sep 2026 13:31:55 GMT<![CDATA[Reply to This is so relevant. on Tue, 29 Sep 2026 12:25:10 GMT]]>@Remittancegirl 100%. I had this with a colleague who sent me 'the right info to set up something' in a mail server. From crapgpt.
After the first 4 words I already knew it was wrong, which makes me distrust anything ai since then.

]]>https://forum.fedi.dk/post/https://writing.exchange/users/paulk/statuses/117354313189032087https://forum.fedi.dk/post/https://writing.exchange/users/paulk/statuses/117354313189032087Tue, 29 Sep 2026 12:25:10 GMT