This is the direct result of people like Sam Altman lying through their teeth about what LLMs can do, but obviously there's a long history of lying about what deep learning-style 'AI' can do.
You can see in it the overwhelming desire for these people to be at the 'forefront' of the next 'revolution'.
Dude, all of this stuff - deep learning, LLMs, etc. etc. - it just takes a corpus of HUMAN GENERATED CONTENT and then tries to make inferences with ZERO UNDERSTANDING OF THE CONTENT.
It's fucking overhyped slightly fancier regression. Autocomplete on steroids but infinitely less reliable because it's allowed to go off and live in its own black box.
It's total boomer bait for people who don't do any research or engage their brain in any way, which sadly means most of the world's media and incredibly institutions like the IMF.
It's depressing, like a mass delusion right in front of your eyes.
https://www.bbc.co.uk/news/business-67977967
0
0
3
Petr Tesarik
ptesarik@fosstodon.org@agilegeoff @ljs Yes, there are some valid use cases for LLMs. The issue is that with such limited scope, the exorbitant costs of running a LLM instance would be quickly rejected by businesses. So, LLM providers must hype up the thing quite a bit to appear competitive.
1
0
1
Read https://blog.piekniewski.info/2023/04/09/ai-reflections/ for a more thoughtful rebuttal.
1
0
1
Cost and CO2 and copyright theft are all added 'bonuses' to this scam.
You can't infer your way to dynamics. It's a black box.
Again I'm too busy this morning to debate it but https://blog.piekniewski.info/2023/04/09/ai-reflections/ for a more thoughtful exploration as to my point.
1
0
1
Which it absolutely must do due to the technique, is inescapable, and unfixable. The 'life coach' one is borderline evil by the way, think about it.
You can't infer your way to dynamics.
0
0
1
Petr Tesarik
ptesarik@fosstodon.org@ljs @agilegeoff Yes, part of the issue is that LLM providers have successfully externalized their costs. But it's too expensive for what it can do even if you consider only the admitted costs.
1
0
1
Read the article. Also read https://en.wikipedia.org/wiki/Eliza_effect?useskin=vector
I'm not trying to be dismissive and arrogant here btw I have a medical appointment shortly and don't have time to go into detail
1
0
0
Petr Tesarik
ptesarik@fosstodon.org@ljs @agilegeoff You say: “But it can't do what is claimed of it.” I can't agree more. It's basically what I'm saying.
IIUC @agilegeoff argues that LLMs can still do at least *something*. And that's somewhat true (although the provided examples seem to be wrong). My point is that this “something” that AI can actually do is not worth it.
1
0
0
Pavel Machek
pavel
1
0
1
Petr Tesarik
ptesarik@fosstodon.org@pavel @agilegeoff @ljs I doubt AI is worth replacing humans. Humans are surprisingly efficient for many tasks…
0
0
1
You're showing that you're a typical AI fanboy type. You've ignored all the points put to you, all the nuance, the article, the etc. and you've literally misrepresented an article to promote your obvious position.
Also who is saying you _can't_ get a hallucinating copyright theft machine to write articles? You can, it's just a stupid idea and won't work in the long run.
What do you think the AI uses as sources? Who writes them? Think about it.
Anyway you're not in good faith, I gave you a chance for reasonable debate and you're showing your true colours. So block I think!
Also can't help but notice you're an agile advocate... let's just say 'not surprised' and move on...
0
0
1
Petr Tesarik
ptesarik@fosstodon.org@agilegeoff @pavel No, that's not what I mean. If you're unable to pay attention to anything without a buzzword, read about “model collapse”. Because that's at least somewhat close to my angle.
0
0
0