@thomasfuchs That "I always check" has the same energy as "It's open source so I can audit the code"
Conversation
Notices
-
Embed this notice
mr_daemon (mr_daemon@untrusted.website)'s status on Tuesday, 02-Jul-2024 00:55:15 JST mr_daemon -
Embed this notice
isaiah (isaiah@mastodon.social)'s status on Tuesday, 02-Jul-2024 03:01:22 JST isaiah @thomasfuchs i don’t use LLM much for anything so i don’t really have a strong opinion on the matter.
but i’m just curious about this point…
LLM are potentially faulty. And worse they seem very confident while lying as they have no idea if what they’re saying is correct.
But something similar can be said of StackOverflow and other crowd sourced info. Right?
And to some degree even my own memory sometimes lies to me and tries to convince me it knows something it doesn’t. 😂
-
Embed this notice
isaiah (isaiah@mastodon.social)'s status on Tuesday, 02-Jul-2024 03:01:22 JST isaiah @thomasfuchs i don’t need you to respond, by the way. i was just scrolling past.
i fully agree with your point, of course.
and i feel the same about SO — which i only use sparingly when i get very stuck, and would never consider copying and pasting from.
but i know im in the minority. i think there are a great many developers that copy and paste code which they only superficially understand.
i don’t really think that’s a good thing. but it definitely a thing. many are doing it.
-
Embed this notice
Brahn (brahn@hachyderm.io)'s status on Tuesday, 02-Jul-2024 10:17:50 JST Brahn @thomasfuchs lol. "i know better than everyone"
-
Embed this notice
spielhoelle (spielhoelle@hachyderm.io)'s status on Tuesday, 02-Jul-2024 10:53:35 JST spielhoelle @thomasfuchs it worked for me so well, still after reading this. It’s if used accuracy like for example summarization, expansion, coding support, template writer, finding needle in haystack etc. it’s a free (ollama), offline super powerful daily assistant to me. And we are at the verge those models get better.
-
Embed this notice
Boris Schapira (boostmarks@hachyderm.io)'s status on Tuesday, 02-Jul-2024 20:02:37 JST Boris Schapira @thomasfuchs What hurts is that 'working with others' obeys the same law. You can't tell if they're lying, and they don't even do it intentionally. They're also full of bias and manipulated by others. You can't always check. Trust' is a good way to go, even if disappointment is certain.
I suppose that's why people are so inclined to trust a tool that talks "like a human": it's because trust is a key element of human collaboration.
-
Embed this notice
DocRekd (docrekd@hachyderm.io)'s status on Tuesday, 02-Jul-2024 21:11:47 JST DocRekd @thomasfuchs Just because I double check it doesn't mean that it's double the work, sometimes even a wrong start can help beginning
Besides computing is full of examples of euristhics and nobody call them useless -
Embed this notice
Toon Van de Putte 🍉 (toon@toonvandeputte.be)'s status on Tuesday, 02-Jul-2024 23:13:09 JST Toon Van de Putte 🍉 @thomasfuchs People also use them to write better text, with the information they provide. Lots of people overestimate the quality of text 'written' by AI, though, in my opinion. Lots of platitudes, always the same overly enthusiastic tone of voice and full of fluff.
-
Embed this notice
fsniper (fsniper@mastodon.social)'s status on Wednesday, 03-Jul-2024 03:41:36 JST fsniper @thomasfuchs I get your sentiment. I understand what you mean. However I can't agree on some perspective. We use humans for so many different tasks, we pay them, and they still do mistakes. They lie, they are unpredictable. They are expensive on some contexts. And we don't say they are useless. Everything can be useful given constraints.
-
Embed this notice
sudo_bash (sudo_bash@fosstodon.org)'s status on Wednesday, 03-Jul-2024 03:42:14 JST sudo_bash @thomasfuchs plus, I learn from my mistakes. I don’t want a computer to make the mistakes for me. I don’t want to lose those learning opportunities.
-
Embed this notice