Post AlSKp5yK2s77Lo96Qa by selea@social.linux.pizza
(DIR) More posts by selea@social.linux.pizza
(DIR) Post #AlSBpMz5tVUiWhYxMW by selea@social.linux.pizza
2024-08-29T08:35:53Z
1 likes, 1 repeats
Spent some hours yesterday trying to debug small python-script, that was supposed to run on a Raspberry Pi together with somebody else (not dayjob).The person was stubborn and decided to trust and ask ChatGPT constantly about all the issues, and it just spew out all sorts of nonsense constantly - while all the correct solutions was a web-search away.I dont know what happened with people after they got exposed to chatgpt, but some of them blindly trust everything that it print out for some reason.And it turns out, that they have followed chatGPTs advice and just removed some system components in a previous session,,,And that's why everything on that poor pi was broken
(DIR) Post #AlSCIrdizuA204dDU0 by nozze@social.linux.pizza
2024-08-29T08:41:15Z
0 likes, 0 repeats
@selea I really hate it that people are this gullible. ChatGPT IS NOT SMART, it's just a very large database with data, and a lot of it is shit posts.
(DIR) Post #AlSDDapi9mUMNiHwwq by rollspelosofen@mastodon.nu
2024-08-29T08:51:26Z
1 likes, 0 repeats
@selea I've used ChatGPT once or twice like that and you need to be alert while doing it, to see through all the bullshit that it provides.
(DIR) Post #AlSDhToio3iXJF4pAO by selea@social.linux.pizza
2024-08-29T08:56:53Z
1 likes, 0 repeats
@rollspelosofen Agreed.You need to know atleast the "advanced-basics" in a subject before you can actually ask it stuff. I notice way to many small errors all over
(DIR) Post #AlSHaM0npKu0tLaLM8 by aaron@social.caskey-demaret.se
2024-08-29T09:40:25Z
0 likes, 0 repeats
@selea A slightly different rant: I don't even care if it gives the correct answers, the environmental cost of LLMs is just too damn high. The fact that it's terrible at the one thing it's meant to do just makes it more wasteful.
(DIR) Post #AlSIFfkPM65FyPcQUq by TFG@social.linux.pizza
2024-08-29T09:47:54Z
0 likes, 0 repeats
@selea That's what I experience almost every day at work. But not with ShitGPT.We provide viewing software to our investigators to view digital evidence. And we ALWAYS provide a custom written manual how to use it (just 2-3 pdf pages) which covers EVERYTHING they have to know to do their work perfectly.This simple manual is ignored in 90%+ every time. Instead they take the phone and call us to let us explain what they could have read instead. 100% waste of work time.People don't want to do the work their-self. People don't even want to use their brain in some cases.All they want is someone who does all the things for them instead.It's so frustrating. That's why I really like pages likehttps://googlethatforyou.com?q=social%20linux%20pizza%20on a private basis. For people like those mentioned above.Frus-tra-ting.
(DIR) Post #AlSKaPudcmScqrQJLU by thomy2000@fosstodon.org
2024-08-29T10:14:01Z
0 likes, 0 repeats
@selea When you're helping a person and they turn to LLMs for answers, that's your cue to leave as they clearly (think they) know better. I just wish them a fake good luck and leave them be.
(DIR) Post #AlSKlPVNTnUGm0qleq by selea@social.linux.pizza
2024-08-29T10:15:55Z
0 likes, 0 repeats
@thomy2000 Never thought about it that was honestly,
(DIR) Post #AlSKp5yK2s77Lo96Qa by selea@social.linux.pizza
2024-08-29T10:16:44Z
0 likes, 0 repeats
@aaron LLM is a waste, really. I does not provide any value
(DIR) Post #AlSLxQXrXRB7QcJ38y by brunofontes@fosstodon.org
2024-08-29T10:29:21Z
0 likes, 0 repeats
@selea I don't know why, but people tend to ignore any instruction (even if they were very clear and complete), to ask for any random person who could provide a generic answer. We even had that "let me google that for you".Now that random person became the LLM...
(DIR) Post #AlSsmQ9Xx9jZVvIB1s by chebra@mstdn.io
2024-08-29T16:36:04Z
1 likes, 0 repeats
@selea Maybe ChatGPT will actually create more jobs...Seriously though, I was explaining something based on my 15 years of experience, and one guy told me "that's not what ChatGPT is telling me but ok" and I almost lost it.
(DIR) Post #AlTRBITh1TM3OJkUfQ by AkiqhrnGcZEYc1Png8.eltheanine@the.teabag.ninja
2024-08-29T23:02:29Z
0 likes, 0 repeats
That's just.... stupid.
(DIR) Post #AlU4PgnZICjB5Qto2q by selea@social.linux.pizza
2024-08-30T06:22:13Z
0 likes, 0 repeats
However,I solved all that shit without any AI and those fancy tools.MANPAGES
(DIR) Post #AlU72A1LIFwid7oXUu by momo@social.linux.pizza
2024-08-30T06:51:38Z
0 likes, 0 repeats
@seleaHow is it that people switch their brains off? I mean even with solutions from the internet I tend to test them in an isolated script to make sure it works as intended. But if someone is like "You need to uninstall your glibc, this always make problems" I will at least be like "Okay, before I do this, what uses that package and will it probably break when I do this?"(Kids, don't uninstall your glibc! I mean it! That was an example, because basically EVERYTHING needs glibc as dependency to run.)
(DIR) Post #AlU7Ag1cBZtOCs0DE8 by steeznson@glasgow.social
2024-08-30T06:53:07Z
0 likes, 0 repeats
@selea Last time I tried to use ChatGPT for help writing a custom objectMapper Strategy in Java. It lost me an afternoon because it confidently told me the complete inverse of the implementation that I was supposed to use. Progress only came when I got frustrated and started again without any AI assistance.
(DIR) Post #AlU9n5mZ4LGt0IPNpo by joakimfors@mastodon.green
2024-08-30T07:22:29Z
1 likes, 0 repeats
@selea Heh, no shit. I always do a facepalm when people ask an LLM which spews some crap instead of going to the project page and looking at the docs/examples. The LLM answer is usually just a mangled version of the documentation with some of the most insane answers from stack overflow mixed in.
(DIR) Post #AlUBOmUtJomC9Og2vQ by antsu@p.antsu.net
2024-08-30T07:40:32Z
0 likes, 0 repeats
@selea "Did you use AI??""No, I used RTFM."
(DIR) Post #AlUDyxyiJGSS5k0ag4 by snukey@mastodon.social
2024-08-30T08:00:29Z
1 likes, 0 repeats
@joakimfors @selea sometimes I use it for smaller questions. It's a bit like Websearch 2.0 but without all the noise of people word fighting in forums. But I don't trust the answers but it can give a direction/starting point at least. I have used Perplexity a few times during js13k to rewrite a messy, but working, loop into a bit more structure. Sometimes it works, sometimes not.
(DIR) Post #AlUDyypBACQQiSYUSG by selea@social.linux.pizza
2024-08-30T08:09:28Z
0 likes, 0 repeats
@snukey I agree that it can be used for some tasks, but blindly trust the answers it gives is wrong.@joakimfors
(DIR) Post #AlULqFd27GYbrODsMS by joakimfors@mastodon.green
2024-08-30T09:37:26Z
0 likes, 0 repeats
@selea @snukey Yeah, it's possible to use it as an advanced fuzzy form of autocomplete or auto refactor but you need to know all the stuff yourself so you can spot the hallucinations. The problem is that a lot of people use it instead of educating themselves.... there are also no nitpickers around to shit on the answer (right or wrong) like on SO so no "meta-hint" if the answer is on the right track unless you yourself have the knowledge. :P
(DIR) Post #AlUhNEb54Pt3FqsX2G by swope@mstdn.plus
2024-08-30T13:38:49Z
0 likes, 0 repeats
@selea I'm no psychologist, but my hunch is there's something akin to pareidolia (e.g. seeing faces in non-face objects) but for intelligence.Like the human mind has heuristics tuned to recognizing intelligent behavior, and we are much more prone to overestimate intelligence than underestimate.Chatbots are designed to create that illusion and take advantage of those heuristics.