ah yes, using a highly specialized AI intended for image generation to create something that is usually in text form. truly a good judge of the quality of AI…
Who said literally anything about Dall-e’s quality? We’re just laughing at the results because they’re funny.
I haven’t been able to get it to print labels on each thing like that, but dall-e definitely has difficulty spelling, or even using real letters. Sometimes when I don’t even ask for text it just prints part of the prompt somewhere in the image in the style of a live laugh love decal, but with at least one word spelled wrong or corrupted.
Nah it’s really bad at labeled diagrams. It knows what words to put on, and now to make it look good, but it doesn’t know how to actually map things well.
It’s really funny but I share your skepticism. At least the labels appear to be edited in after the fact, they’re way too clear and similar and not messy enough to be real, unless there’s something new with DallE-3 I guess.
Why is it even called artificial intelligence, when it´s obviously just mindless artificial pattern reproduction?
Machine Learning is such a better name. It describes what is happening - a machine is learning to do some specific thing. In this case to take text and output pictures… It’s limited by what it learned from. It learned from arrays of numbers representing colours of pixels, and from strings of text. It doesn’t know what that text means, it just knows how to translate it into arrays of numbers… There is no intelligence, only limited learning.
Machine Learning isn’t a good name for these services because they aren’t learning. You don’t teach them by interacting with them. The developers did the teaching and the machine did the learning before you ever opened the browser window. You’re interacting with the result of learning, not with the learning.
Are we so different?
Isn’t meaning just comparing and contracting similarly learned patterns against each other and saying “this is not all of those other things”.?
The closer you scrutinize meaning the fuzzier it gets. Linguistically at least, though now that I think about it I suppose the same holds true in science as well.
Yes, we absolutely are different. Okay, maybe if you really boil down every little process our brains do there are similarities, we do also do pattern recognition, yes. But that isn’t all we do, or all ML systems do, either. I think you’re selling yourself short if you think you’re just recognising patterns!
The simplest difference between us and ML systems was pointed out by another commenter - they are trained on a dataset and then they remain static. We constantly re-evaluate old information, take in new information, and formulate new thoughts and change our minds.
We are able to perceive in ways that computers just can’t - they can’t understand what a smell is because they cannot smell, they can’t understand what it is to see in the way that we do because when they process images it is exactly the same to a computer as processing any other series of numbers. They do not have abstract concepts to relate recognised patterns to. Generative AI is unable to be truly creative in the way that we can, because it doesn’t have an imagination, it is replicating based on its inputs. Although, again, people on the internet love to say “that’s what artists do”, I think it’s pretty obvious that we wouldn’t have art in the way we do today if that was true… We would still be painting on the walls of caves.
Well, I think it comes down to a fundamental belief on consciousness. If you’re non religious, you probably think that consciousness is a purely biological and understandable process. This is complete understandable and should be replicable. Therefore, artificial intelligence. But it’s hard as dong to do well.
Because that’s what intelligence is. There’s a very funny video floating around of a squirrel repeatedly trying to bury an acorn in a dog’s fur and completely failing to understand why it’s not working. Now sure, a squirrel is not the smartest animal in the world, but it does have some intelligence, and yet there it is just mindlessly reproducing a pattern in the wrong context. Maybe you’re thinking that humans aren’t like that, that we make decisions by actually thinking through our actions and their consequences instead of just repeating learned patterns. I put it to you that if that were the case, we wouldn’t still be dealing with the same problems that have been plaguing us for millennia.
If you’re interested in a more detailed exploration of this topic than an online forum post consisting of a grand total of five sentences can contain, feel free to read a book.
AI is also the minmax algorithm for solving tic-tac-toe, and the ghosts that chase Pac-Man around. It’s a broad term. It doesn’t always have to mean “mindblowing super-intelligence that surpasses humans in every conceivable way”. So it makes mistakes - therefore it’s not “intelligent” in some way?
A lot of the latest thought in cognitive science couches human cognition in similar terms to pattern recognition - some of the latest theories are known as “predictive processing”, “embodied cognition”, and “4E cognition” if you want to look them up.
Some tech bro will attempt to make this cake and will tell someone it was better than anything some uppity WOKE human baker could have made, regardless of how bad it turned out.
Every time I see posts like this I remember a frequent argument I had in the early 2000’s.
Every time I talked with photography students (I worked at an art school) or a general photography enthusiast, I got the same smug predictions about digital photography. The resolution sucked, the color sucked, the artist doesn’t have enough control, etc. They all assured me that digital photography might be nice for casual vacation photos and maybe a few specialty applications but no way, no how, not even when hell freezes over would any serious photographer ever consider digital.
At the time I would think back to my annoying grade school discussions with teachers who assured me that (dot matrix) printers just sucked. Serious writing was done by hand and if you didn’t know cursive you might as well be illiterate.
For some reasons people keep forgetting that technology marches on. The dumb glitches that are so easy to make fun of now, will get addressed. There are billions of dollars pouring into AI development. Every major company and country is developing them. The pay rates for AI developer jobs attract huge amounts of people to solve those problems.
There’s either the “it’ll never work” take or the “it’ll destroy the industry!” take, and both are kinda childish. New technologies are tools, nothing more, nothing less. Learn to use them and they’ll make your life easier. Integrate them if they’re threatening your livelihood. Learn and adapt, it’s how progress has always worked.
And up to now we have zero indication that the current approach isn’t a dead end. Bill Gates, for instance, thinks that GPT-4 is a development plateau: https://heise.de/-9337989
Bill has made some famously bad predictions in the past. Here’s a small sample https://www.independent.co.uk/news/people/the-worst-things-bill-gates-ever-said-a6990046.html
It’s possible that the current $100 billion market size of AI and all the AI job openings are completely misplaced but that’s indication that a lot of people have pretty high expectations that AI will continue to grow.
Ah, yes, famous expert in artificial intelligence and machine learning, Bill Gates. I’m personally curious what Taylor Swift thinks about Chat GPT 5, myself. That girl’s got a lot of money, which means she must be smart and has smart opinions on topics like generative AI and the efficacy of currently undeveloped LLMs.
Same. Remember the same arguments. Heck I still get into it with clients sometimes. Usually snark works
Me: wasn’t 2013 nice? I had a full set of hair and didn’t have to diet, but as much as I might miss 2013 it isn’t 2013 anymore. Time to move forward.
I’m guessing this argument has been going on longer than either of us can remember.
There was a long time when guns were considered interesting toys but not something a sane person would take onto the battlefield; especially not without some sort of backup. Hell, the “three musketeers” were more known for their fencing than their firearms skill.
I’m sure back in the day some chucklehead complained that papyrus was cute but anything important had to be carved in stone tablet.
Certainly, but none of those technologies completely replaced things. The existing way of doing things became hobbies and remain the preference over the technology which disrupted the field.
Not to mention, technologies will sometimes flop, only to resurface later in a completely different package. The PDA was maybe popular for a year? But now we all have smartphones which effectively capture that concept. The Wii U failed, but the Switch has been wildly popular.
It’s probably premature to say that AI will completely fail, but also that AI will completely replace everything. I just used a Polaroid camera this past weekend at a wedding, and it was enjoyable in a way digital cameras or phones wouldn’t have been. I still write things out at work, particularly if I’m trying to wrap my head around some math or a difficult concept. Typing it out doesn’t work as well.
I think it is safe to say that there are some things AI will never be able to replace, just like there are some things digital cameras couldn’t replace, nor our phones.
My wife still likes film photography. We’re even thinking of setting up a dark room in the basement. Turns out it’s a huge PITA to find someone who knows how to repair a Mamya and, despite her best intentions, every photo she’s taken in the last decade has been digital.
As near as I can tell film photography and hand-written letters are things people do for fun. Both digital photography and modern printers can produce outputs that are are nearly indistinguishable from film or handwriting. I’ve been at weddings with Polaroids too. It’s fun but the pictures aren’t very good. The color is off, the resolution is low and they fade faster. The only advantage is that everyone gets to do a fun retro-thing. There’s nothing wrong with that, I have a lot of hobbies that involve doing things the old fashion way just because it’s more fun.
When we actually need the results to work well we generally go for the most modern technologies we can afford. When the main purpose is the enjoyment that accrues to the creator, there’s still a lot of room for older technologies.
That’s what I expect AI will eventually get to. Right now AI art is largely a novelty. Soon it will be the standard and creative artists will find interesting things to do with it that none of us are considering now.
