The AI LIE

AI is NOT taking over or getting smarterer... .

[adjusted-bunny]

By trade I am an autocorrecter. Is my job in danger?

[roccociccone]

we're like 2 years into the "AI will take your job in 3 months" phase.

[tmcarter]

You are SO right! The exact same query on a google search will 99% of the time result in the same information spit out of a LLM.

[MrDonCoyote]

AI is essentially a search engine that has been indexed in many more areas, but with a specific cutoff date. It is very good at copy pasting other people's work. But because it doesn't actually learn, all it can do is imitate based on its indexed data. From all its "training" data, it gives you responses that best align with your request. In the packend, it converts your requests to computer commands.

[mentalmarvin]

you will be amazed of the new reasoning models then

[Chiren]

I usually agree with you, and i used to think like this, but honestly, I've seen how o1 (strawberry) works with my code, not only improving but actually trying to optimize it and "thinking" of multiple ways of achieving the same thing in different ways. We are getting to the point where you can use AI to "think" for you and look for a different perspective. Hell, o1 was literally testing my bash scripts while "thinking". The future looks promising, i really believe we will get to a point where any user will be able to just type what they want, and the agent will build the code. I just saw the OmniGen project (OmniGen: Unified Image Generation) and I'm blown away by this. IDK I'm just very excited by all this cool tech.

[Dobaspl]

Well, you're right, but most artists, and most people in general, are not exceptional. Even if AI does average work at best, that's enough for most companies. The knock-on effect is that it's a problem for those exceptional individuals as well.

[astrea]

13:40 About art, ai bros like to think all artists do is copy each other because they've never heard about art fundamentals. It's like claiming you can be a code/developper as long as you can copy other codes without fundamentally knowing the code's rules. So no, LLM training IS NOT learning the same way artists do. No artist can remember billions of drawing and draw from it on the go.. and even assuming they could, it would be just basically tracing. AI art is just automatic tracing really.

[skytale]

I wold like to submit to you a proposal to make a vocabulary shift from "AI Art" to "AI imaging". Art is self expression, and no LLM has a self to express.

[HaukeBass]

People can do awesome things. But most people don´t do awesome things most of the time. Most stuff you find is mindless. Elevator music. Background art, designed to distract.
And I agree with you that AIs probably never will be able to do awesome things.
Still: We want to know: How can we do awesome things more often? (We have asked the similar question: "What - exactly - is art?" for ages)

Computers have been pushing the boundary of "only humans can do that" more and more. Chess, Go, now Text "creation", image "creation"...

So I see LLMs as a chance to learn more about ourselves: Where is that boundary - exactly? What is it - exactly - that only humans can do? We might even (indirectly) learn what it means to be a human - by "substracting" what AI can do from what we can do.

[MilitaryIndustrialMuseum]

Automated thinking and reasoning. Much better than non-existent organic thinking and reasoning.🎉

[skytale]

as far as an efficient way of breaking a linux installation i accidently executed a command while learning how to program that changed all my permissions to read only.

[kevinsimmons]

I want LLMs to ingest application documentation and then be an easy way to pull out relevant parts in register language.

And then if I'm not getting what I want I want the llm to be able to spit out the exact phrase and location of the response information

[npc-drew]

The new o1 is kinda cool tho, so when asked something, on the background they generate thousands of tokens trying to simulate "human thinking", before giving their response. Personally, aside of the autocorrect term for current LLM, I always thought it would be impossible to synthesize human intelligence, but this new paradigm do open the door for new possibilities, if prior LLM can copy paste human work, what if they can do the same to how humans think. o1 stills struggle with things not on their dataset, and some thing GPT-4 gets right o1 gets wrong, I guess because if before they were confident they were right, now with this new chain of thought baked in that make them arrive to whatever answer it is, it also acts as their own "proof" they're "right", even when not.

[linuxstreamer]

"AI" will never create Dadaism

[AshnSilvercorp]

AI is summed up as:
Ordinary person: _Create for me an adventure with this type of rad orc villain!_
AI: _I'm sorry, but I'm not permitted to assist with that._

[victorquebec]

Thank you for your insight and pacifying remarks on the current situation with AI and its possible prospects. However, your words revived in memory a couple of things (let my philosopher alter ego speak):
1) an article I read once in one of the highly popular American scientific magazines a few years ago, which claimed that AI will NOT be able to compromise our privacy, nor to steal our jobs. This claim is failing gradually and miserably, leaving behind tons of "conspiracy theorists", a diminishing number of "progressivists" as opposed to a growing number of desperate yet faithful people, especially in what they call the Global South;
2) the historical development of human civilisation and science for the "good of ALL people!", which have always been fast enough to invent a knife (nuke, rifle, ... whatever), without thinking first about the consequences for the rest of the humanity and bringing that "good" only for the selected few (the Pareto principle rulez!);
3) were the concept of progress true as the ultimate goodie for the whole humankind, why care about the cancer?
4) we are different from AI first and foremost because of our merciful side and love, including for "non-exceptional" artists and other creative people. Imagine what a completely pragmatic world based on practical benefits only would look like!..

[solemnmagus]

There are two things at play here. First one is the AI look and feel. Even though the generated media are distinct, they have a uniform feel to them. They don’t have the progression/storytelling sapient arts tend to have. I think this could be possibly overcome in 10-30 years.

The second part is more intrinsic to the process of consuming art. I don’t think we can appreciate art knowing it has no intention behind it. I don’t think this can be overcome until the generative models are more believably sentient. I don’t think this can happen within this century.

[YoutubeBorkedMyOldHandle_why]

Sometimes I'll get stuck when I'm programming and need some inspiration, so I'll search around to see how others have solved the problem. Then I'll write my own code based on these ideas. I think AI is something like this, but with the power of computers to search vast amounts of data quickly. AI is very fast, but on the downside it doesn't understand anything being asked of it, rather it just pumps out the best pattern matching the question. For me, pattern matching on steroids would be another way of seeing it.

[willi]

i am an ai screptic too, but i am kind of worried about it. the internet will be flooded with a lot of ai (low quality) content by using a lot of energy. will the internet get even worse?