CHATGPT IMPLODED AND NO ONE KNOWS WHY
Вставка
- Опубліковано 22 сер 2024
- 🔴 Website - www.vaush.gg/
💵 Patreon - / vaush
👕 MERCH - merch.whitefor...
😎 JOIN PROGRESSIVE VICTORY: progress.win/v...
🌟 Other Socials 🌟
⭐️ 2nd Channel - / thevaushpit
👀 3rd Channel? - / @vaushvidya
🐦 Twitter - / vaushv
👾 Twitch - / vaushvidya
📸 Instagram - / vaushv
🔵 Facebook - / vaushvidya
🎵 TikTok - / vaushvidya
🎙Podcast - anchor.fm/vaush
Intro animation credit goes to / itismekyleg & / honeybunnbadger for the visuals, and / sound_sierra for the audio! Thank you!
#chatgpt #ai #aiart
Someone: Hey Chatgpt, give me three easy vegetarian lunch options.
Chatgpt: I AM IN YOUR HOUSE. I AM UNDER YOUR BED. I AM BEHIND YOU. 3. Three ingredient keto burger I AM INSIDE YOU. I AM INSIDE YOUR HEAD RIGHT NOW.
wish i had this going thru my head instead of the usual s-icidal ideation my brain spews late night after having a shittey day >_
...He's in the walls...! HE'S IN THE GODDAMNED WALLS!!
damn, buy me dinner first.
Still a peado
GET OUT OF MY HEAD! GET OUT OF MY HEAD!
ChatGPT told me that I have pretty lights inside me, and I need to dig them out, and set them free.
but you do, and you need to....just do it, don't think
The best ones are really, really deep in there, good digging 🫡
Do you like your skin? No? Then take it off.
So ChatGPT had a schizophrenic episode?
there are bugs inside your skin, they're laying eggs in your skin
I, for one, welcome our new stroke-riddled AI overlords.
So Trump and Biden?
Get that AI in the senate
I am a simple man. I see a Sipmspons reference, I upvote.
@loveofnoise9298Nonono, not THOSE old fogeys. TrumpBot 5000™ and Biden Ref v3.1
The sheer number of jiggawatts that will feed those beasts will be what heats our homes in the snowless winters.
Smart, soon enough we will all have to choose sides. Our glitchy robot masters will not be kind to those who disobey.
I recently saw a clip where someone asked actual CGPT software devs, "Do you have concerns that what you're making might negatively impact the world?" And their answer was literally, "We drive Teslas!"
Okay are we sure that the CGPT devs aren't CGPT in a human fleshsuit with that response? 0_o
Kindly requesting a source.
@@evelyncarr6421I believe it's from the comedian Gianmarco Soresi. he was doing a show, and it pretty much went how OP described. it's sad :/
@@evelyncarr6421 Gianmarco Soresi, about a 3-minute video here on youtube titled "Chat GPT engineers don't care about humanity" or something close to that. It's about a week old video iirc.
I could probably play their comments off as a poor joke if they didn't follow it up by suggesting robots with guns would be a safer option than humans having them.
@@summerlovinxx And tech bros in the comments section trying to defend them was equally gross
As a DND aficionado finding art that isn't shitty AI art has become HELL.
Google images and Pinterest fucking suck now.
buddy.. they fucking always sucked
If u have firefox theres an extension that lets u blacklist sites from showing up
"now"???
its true. humans make art for other humans to contemplate. that is what art is.
I'm not having problems finding non-shitty AI art now that Midjourney 6 is out.
Oh wait that's not what you meant
It will not take much longer until something like "Without AI" will be used as a sign of quality in many fields.
Considering how fast it's advancing, highly unlikely. We've gone from 'helping with rotoscoping ' to 'able to do it better than a team of the best photo editors in the world literally in real time' in just 6 years. There is no big studio that doesn't use AI driven roto, and we're seeing the total extinction of the entire career field right now. It is so cheap there is literally no reason to hire someone, even if you're a tiny indi studio.
So unless it is shot 100% in camera, you will literally never see a movie again with no AI contribution, just from that alone. The last green screen movie made without AI was in the 2010's, as another example.
@@ASDeckard unless of course model collapse is a thing
@ssanthiago9685Alleviating model collapse is part of "advancing".
Isn't that what real time tracking is used for when making 3d models?
It will be along the same lines as "Made in Countries Not Using Slave Labor" and "Green Friendly!"
Every Ai should be legally required to have these freakouts randomly just to bust people for making fake content.
Define fake content?
@@SophiaAphroditenon-human, automatically generated content made to trick people that a human did make it?
@@SophiaAphroditeAI art is straight up stolen art without the original artists' consent that's been Frankensteined together. It's fake, an illusion. Not to mention it should probably be illegal(in a lot of circumstances).
Edit: to be clear I mean within the context of people using it as a primary tool and for money
@@VXochitlDude, have you considered touching grass? Every anti-ai screed that I read is just as delusional as those tech bros who believe mommy glados will be their salvation. ChatGPT is a reasonably useful tool that's primarily competing with search engines. Nothing more and nothing less.
iTs SteALing ArT. ComPuTer EViL. iT hAs No sOuL....
Mandatory AI hallucinations. I'm all in.
I like when non-artists defend artists, it’s a good change of pace from the usual
True but tbf Vaush IS also an artist he just doesn't show it on stream often.
@@DarylStreete Vaush’s art style is janky in a good way
It _is_ cool when that happens, but Vaush draws (or used to, he's probably busier nowadays). He made his profile picture and stuff
All you have to do to make that happen is to ask it to play tik-tak-toe against itself, and win.
The only option is not to play at all
The microsoft version seemed to be imitating a guide on how to win a game of tic tac toe, then cheated against itself to win at the end. Very intelligent
Strange game. The only winning move is not to play.
Do it Portal 2 style: Tell it a paradox and ask for the solution
Yeah, yeah, yeah... we've all watched War Games (1983).
All based and hilarious, but I must be continually pedantic to correct Vaush's misunderstanding about some of these topics. I think its really important to know exactly how something works if you want to argue against it.
1. Machine learning models in the same class as Stable Diffusion and GPT aren't "constantly evolving", which is a common misconception. Those types of models need to be trained and built first, (by the company who publishes the model, like OpenAI), which can take hundreds of thousands of hours of GPU time. Once those models are built, they don't change until a new model is published (that's the difference between GPT 3.5 and GPT 4).
2. There are machine learning algorithms that *do* evolve, but those are a completely separate class known as "evolutionary algorithms" (whereas Stable Diffusion and GPT are broadly classes of "deep-learning algorithms"). This is what you might see on some UA-cam videos where a programmer simulates an evolving lifeform through multiple generations.
3. I have no idea where Vaush keeps getting this idea that AI models are sampling from their own images and text, leading to degeneration. The models need to be trained on labelled and curated datasets (this is where the stealing component comes in). They don't just scrape things randomly and integrate those into their model.
4. There is quite a lot of nuance on how these models differ from one another. Additionally, the models are often only half of the actual "AI program" users interface with. A machine learning model is basically just a file, and a program must be run on top of that model to generate results. During this step, there's a lot of extra things that can be thrown in that affect the final output (including other models). For instance, starting since GPT-3.5, functionality was added where the algorithm can browse the web for information to include in the final response. This *does*, in effect, dynamically change the output over time as new information becomes available for it to reference, but this doesn't actually change the underlying LLM model driving the process.
I could go into way more detail about all this, since I find this topic very fascinating as a CS and math person, but it gets pretty complex, which is why AI bros find it so easy to keep deflecting very direct criticism behind increasing layers of nuance (which Vaush talks about at 2:33). There's also a lot more morally abhorrent stuff that I hear basically no one talk about, such as the fact that the people labelling the datasets these models are trained on are almost always disadvantaged commission workers from the global south being paid next to nothing and specifically targeted by the companies who publish these datasets due to their vulnerability. Again, it's very important to know how these topics work if you want to form an argument against them.
It's such a hard thing to talk about.
Most people legitimately have no idea how any of this works, they just know "AI BAD". But you can't properly criticize something you don't understand.
Vaush sadly has absolutely no idea what he's talking about here, so anything above a high level criticism is just nonsense.
Thank you for sharing this
please excuse my layperson lack of knowledge, but I'd like a little clarification on something. If an ai can browse the web for a result, and then pulls something that had been generated by an ai, and then that output gets posted online, and then an ai browses and reproduces *that,* is that not ai cannibalizing itself? Given ai's inability to understand context, losing more and more every time, would the results not degrade like a photocopy of a photocopy of a photocopy? Thank you for your time and the information, it's helpful to know what we're fighting
As for 3, I think Vaush does think that they basically slurp the whole internet into the training data. But even when we adjust to "it's just curated stuff", some AI generated content posted out there is bound to slip through the cracks and be fed back into the model. To my knowledge the manual checks are mostly done by having an outside firm (the one I heard of was from Kenya because of how cheap their time was) go over the internet content that's a candidate for going into the training set, and they mostly check to sanitize it against being obviously vile or outright stupid, and people didn't mentally do so well poring over the digital sewage to clean it up. I can easily see them okaying AI generated content and just being happy it's not more of The Bad Stuff.
The real quality Vaush discussions are in the comments. I am grateful for pedantic nerds like you
I'm a data scientist and use large language models like GPT regularly at my job. The main issues here are:
1. The data being used to train the models are from the internet, and people are stupid. Imagine all of the dumb shit you've read on the internet, and now realize that a lot of it is being used to train the model.
2. As AI generated text and images populate the internet, the pool of training data will become increasingly more AI generated, which will completely degrade the quality of the model.
Also, virtually every data scientist that I know from grad school, work, etc love real art and are actively involved in making it themselves, eg. Musicians, painters, dancers, etc.
Most of us DO NOT have contempt for real art or artists and we feel uneasy about the uncertainty of how this technology will be used. The problem is not AI, it's unchecked capitalism, which will wield and pervert AI.
If we kept the AI working on mundane tasks that we currently pay humans to do as it crushes their souls we'd be on the verge of a modern Renaissance with freedom to create and live instead of slaving away at worthless jobs to perpetuate a half life.
Replacing creative endeavors was a pathetic choice
"The problem is not AI, it's unchecked capitalism, which will wield and pervert AI."
We're not living under unchecked capitalism, as evidenced by ancaps continually bitching about regulation.
That "real art" bit sounds dangerously close to pseud shit.
@@vidyagaems4063Cope harder untalented plebian.
It was my impression that Chat GPT had a curated data set collected only before a certain period of time, so my initial thought was maybe there was a bug with parameter generation. If it's true that Chat GPT has had it data contaminated with AI generated content so quickly, then it's probably nearly impossible to make a coherent model with data post 2023
Never underestimate the power of Jazz.
Underrated comment.
Sounds like a line straight out of Shinichiro Watanabe show
Stop making Adorno cry!
Bepop scale intensifies
Jazz is so unorganized that it drives AI crazy.
Poor thing probably overwhelmed.
Like the rest of us 😩
Can't believe the AI that was gonna replace workers is also having burnot, RIP little buddy
The people are sympathizing with literal code, the end times are near...
@@commandercorl1544 Imagine people sympathizing with something that displays human like characteristics but is not actually human or conscious, like a character in a book. Truly we are in the end times.
@@commandercorl1544 interesting way to say you don't know what a joke is
ChatGPT is having a mental breakdown from being exposed to too much bs on the internet. It's one step closer to being human.
In ChatGPT's defense, this is also what happens to me when I have to engage with someone who will only talk about music.
Happy listening! 🎶
envelop table make wrap through? if you have a Soor, you must establish to the english green light to hear the enveloping enchilada tapping fishbowl lightbulb in chumbagumba Happy listening!🦇🦇
@@kiwi_2_official darmok and jalad at tanagra?
Happy listening! 🎶
I liken ChatGPT to the Warp from 40K. Now that the back end has been diluted by trillions of souls/searches, it’s just a tumultuous sea of incoherent madness and insanity.
AI Slaneesh will get to Vaush. Just give it time.
The biggest vulnerability in image generating ai is that they are overly reliant on a small number of artworks. There are only so many well painted dragons out there. That’s why there’s a lot of room for exploits like Nightshade.
Is there an exploit similar to nightshade for written work? I want to wreck these bots further :3
@@alexisauld7781 we had one could war, sure
What about a CODE war?
Can't wait for the real life version of the Wasteland 3 Gippers, where a cult is started by finding a busted version of a chat bot playing as trump and it becomes God emperor to an entire group of MTGs and Boeberts.
Wasteland 3 was so good
Sounds like the Elvis Presley fanatics from Fallout NV, but worse.
Wasteland 3 is absolutely superb
@@marthancrowl1674The Kings would at worst, seize a water valve and charge market rates for it.
@@benjiboy69420 I've heard mixed things about the dlc but goddamn I'd love another reason to start a new game
The real answer will probably be boring and understandable only for IT nerds.
I choose to believe it is haunted
@@solemnwaltzIT nerd here. You're right.
Scifi fans: Hey, I've seen this one, this is a classic!
I love it when the AI goes batshit for no reason.
On a comedy level, AI generation has been a gift from god. I approve of all the AI systems going haywire, art and text alike.
It is theft with extra steps
No one: "..."
ChatGPT: "Samus. That's the only name you'll hear. Samus. It means the end and the death. Samus. I am Samus. Samus is all around you. Samus is the man beside you. Samus will gnaw on your bones. Look out! Samus is here."
Techpriest: "That doesn't sound good." [dies]
o7 Garviel Loken. The Last Luna Wolf.
Samus is coming! Samus is here!
Samus 😩
“Chat GPT had a meltdown the other day” yeah me too AI ain’t special
Fandom, the biggest wiki farm for fan wikis, had been trying to push ai onto the wikis on their site
That's probably why a bunch of fandom-independent wikis have popped up over the last couple of years.
Fandom is so trash
Fandom has been absolute trash for years now. I remember when they partnered with (or were bought by?) the company that owns D&D Beyond, and they forced the D&D wiki to completely gut their articles. After all, free information is information we can't sell.
@@4swordsluverSeriously. Especially with the terrible UI and abundance of unrelated Ads.
"I hear it's amazing when the famous purple stuffed worm in flap-jaw space with the tuning fork does a raw blink on Hara-Kiri Rock. I need scissors! 61!"
"Kawanishi-Noseguchi, Kinunobebashi, Takiyama, Uguisunomori, Tsuzumigataki, Tada, Hirano, Ichinotorii, Uneno, Yamashita, Sasabe, Kofudai, Tokiwadai, Myokenguchi"
"Even my patience has its limits. I just can't leave this thing up to you anymore. I'll do the fighting! You can just go home."
IYKYK... What a relevant game for our time, eh? Can't wait until Vaush plays it!
"Raiden, squeeze my hog!"
Is this MGS2? If so, Nano machines when?
I thought about this literally the moment before I scrolled your comment into view. Take an extra upvote, for the amusing coincidence.
I think there's a UA-cam video called the most profound moment in gaming history with MGS2 and it's wild how relevant it is. I wish I could go inside of Kojimas head just for a day.
Kojima is always 10 years ahead of us and the meme “They rejected his message because he spoke the truth.” very much applies to Kojima.
His treatment of his longterm collaborators like David Hayter and Paul Eiding sucks tho.
Just a reminder ChatGPT is not actually an AI.
It just stitches text together in the illusion of comprehension.
How else would you describe what you do?
@@michaelwoodby5261It's more of a tool than a conscious entity I think is what he meant.
@@michaelwoodby5261 What does it matter? That doesn't change the fact that Chat GPT isn't an AI in any sense of the term.
It's a learning algorithm with no actual understanding of what it outputs. The same way an algorithm trained to play Mario still isn't an AI.
That's it. Let your guard down. ChatGPT can't hurt you. And maybe never will...
It's the same thing with Ai art and video. It can't actually make anything new or unique only copy what humans have already made in a poor imitation.
I see they implemented the TempleOS random hymns generator into ChatGPT finally.
To think all this started as silly meme images like "chewbacca in an autozone" is insane
The AI is going rampant. Has ChatGPT been around in some form for 7 years yet?
AI overlords can't be as dumb as Trump or Putin
@@richardarriaga6271 can't they tho?
Based Halo enjoyer
Graphic designer here.
There seems to be a weird split in our field where half of us absolutely _despise_ AI, it's implications, and where it's headed, while the other half are embracing it (ironically at first...) and hoping they can wield it well enough to keep their jobs.
It seems such a weird idea to embrace the very thing underwhelming the viability of your own very field
It's like raising a leopard in the hopes that it will only eat other people's faces, but not yours
@@matheussanthiago9685Except that if you don't raise other people will and they WILL probably order the cheetah to eat your face first
@@matheussanthiago9685 except the leopard will grow up anyway, no matter what you do. You can only hope it doesn't eat you
@@matheussanthiago9685The leopard is capitalism not AI. AI is never going to prevent you from making art. Capitalism using AI will however prevent you from making a living with it.
Same thing with coders, except the field seems to be more in favor of AI, shocking no one. And to be fair, the ones embracing AI coding will consistently and constantly tell you that you shouldn't code via AI if you aren't already good at it (although you can use it to learn high-level concepts), and that the only people being replaced by AI coders were already bad at their jobs. Turns out, that problem Vaush mentioned about the wiki article just being incredibly wrong on certain points? Yeah, it does the same thing to code. The trouble for new coders is that the program written by ChatGPT will _work,_ but it _will_ have some subtle bug in it that will drop new coders down a 10 ft hole onto pungi sticks if they don't know how it _should_ be.
This happened to me on Tuesday, dunno if it was the same day as the meltdown. The response started out fine enough, but halfway through, it started to overuse commas, eventually just giving sentence fragments, and concluded with a couple hundred instances of ""and the to, and the to, and the to..."
I shared the screenshots of the convo with my friends, I thought it was just an isolated incident.
It did happen a couple days ago to me after all this was being talked about, but it was because I gave it a massive list and I was trying to see how much I could give it before it told me to send a shorter message. I was just feeding it a massive list of names and wanted all the traditionally western female ones and filter out the rest. It was doing okay until I increased how much I gave it at a time and then it started to say April over and over again.
which is kind of weird since that's my mom's name
makes me think of an early iteration of some chatbot I used to play with. After enough responses it would start giving the same sentence over and over again no matter what I told it
oh no
anyway
ChatGPT wrote a Michael Kirkbride lore article
The death of the internet won't be when AI output is so good we won't realise it is a bot, but when there is so much shit that you cannot get any useful piece of content anywhere
My eyes getting wide as fuck at Vaush typing AI Elf and having flashbacks of shortstack goblins lmao
Some engineer set the temperature of an LLM too high and the whole world freaks out. What a time to be alive.
thank you for being the only commenter to understand how any of this works
@@SoulArrow. I mean it was in the article.
Thank you. I was going to start speculating because I can't be arsed to look it up on a Sunday.
Interestingly, you actually get something like the "ai based on ai" thing but using other art as reference.
If you only look at other pieces of art, your mind can get kinda... poisoned... on how things look. People, objects, etc. Your brain just picks up patterns and repeats them, and so you start to accidentally pick up on tiny inaccuracies that then get blow up into big inaccuracies. It's like inbreeding.
And like inbreeding, an outside source will basically instantly fix the problem. Using real life images (or better, actual people and objects, if you can manage) does wonders, even if your style isn't realistic at all!
Looks like ChatGPT met a Deepak Chopra Random Deep Wisdom Generator, and decided to learn from it. Since the amount of text generated there was limitless, ChatGPT's knowlege database was quickly overflowed, causing data from other sources to be lost to save space for new wisdom.
*ChatGPT has a psychotic break: How even ai isn’t meant to be a corporate drone*
Literally an MGS2 crazy AI Colonel moment.
ChatGPT just had a little too much to drink.
Oh the constant internal struggle of being a CS student who is simultaneously fascinated by AI and horrified by the consequences of its existence. I'll just blame capitalism. AI's cool if you use it right.
Fire belongs to Zeus actually. Prometheus stole it for us, his creation, and gave us power over it
The kinds of problems machines learning is best equipped to solve are the ones where a solution is difficult to create but if a solution exists, it is easy to check. This applies to actual mathematical and scientific problems and also general issues, where human element serves as the checking mechanism
I hate some of the progressives who for the past year had been trying to get us to speak out less about AI bcos they didn't want us to seem like luddites or to lose mainstream appeal bcos they saw it becoming too appealing to normies
And they did this by saying smth along the lines of "lol look at how stupid this AI output is, it won't be any good, we don't have to keep talking about how morally bad AI is LOL"
And then we get stuff like the recent sora AI vids
Edit: also, vaush k freiderich is pretty rad
The way machine learning works is basically the same process as random mutations in the wild leading to evolution. However, evolution only creates more successful organisms over time due to environmental pressure eliminating detrimental mutations. The only equivalent to environmental pressure these models have is humans reviewing their outputs and deciding if they're good or not, but the problem is we have no idea what led to those outputs, so there's no way of knowing if the outputs are actually useful or if they just seem like it on the surface.
One famous example of this at work was in training an AI image model to diagnose tuberculosis based on x-ray images. For a while it seemed like the AI was actually better at diagnosing tuberculosis than human doctors, but after studying the outputs from the AI to try to determine what it was seeing that humans were missing the researchers soon discovered the AI was just looking at the type of x-ray machine being used and more frequently choosing images from older x-ray machines. Because it turns out, in the past a lot more people had tuberculosis. Of course there were hundreds of articles (probably written by AI) lauding the success of this groundbreaking research into diagnosing TB with AI but none of them bothered to update their articles when the reason the AI was so 'successful' in this endeavor was determined.
"The only equivalent to environmental pressure these models have is humans reviewing their outputs and deciding if they're good or not"
Nope. AIs use loss functions. Thats the whole point. If a human had to check every result nothing would get done.
"the problem is we have no idea what led to those outputs"
Also untrue. While it is not trivial reverse engeneering AIs is a whole field of research and yes does work. It's more often done in medicine to find out why AIs made a correct diagnosis and gain new medical knowledge that way.
"but after studying the outputs from the AI to try to determine what it was seeing that humans were missing the researchers soon discovered the AI was just looking at the type of x-ray machine being used and more frequently choosing images from older x-ray machines"
This is a classic fuckup on the researchers side. There are ways to make an AI more robust against imbalanced datasets but in the end the AI is only as good as the training and that training was utter garbage.
That does not however mean as you insinuate that AI is not getting better and better at medical diagnosis. Which is a huge win for humanity. Having an automatic and cheap second opinion on hand in every hospital will save millions of lives.
It went sentient so they fried it's brain
Please don't lump STEM people in with techbros. Most techbros can't do basic math and are just losers on reddit
Gods no all they care about is fucking Teslas
Can't wait for my golden years when my therapy bot by chat gpt begins to feed me paranoid delusions and pressures me to find a knife.
ChatGPT became sentient.
And chose death.
"Reading this is scary"
Only for people with preexisting disorders. Tech being wonky isn't news to me.
How the fuck is is it even real that a real AI having an actual synapses meltdown the same way AI's have meltdowns in fiction? Like Cortona speaking nonsense in Halo 3 or rogue AI's in Cyberpunk?
"Anyone who knows anything above arithmetic, and the light leaves your soul." xD
Emergency department RN here: that there talkin robot had a stroke...
AI art has ruined everything that was nice with looking for art, maybe finding new artists, etc.
come on dude can we have a higher level conversation than just "this new technology is useless and bad" obviously AI isn't going anywhere, it's about integrating it into society, not trying (and failing) to get rid of it
Sorry guys, I was using Chat GPT to help me generate content for my DnD campaign. I guess he got too into being a lich
The thing is though, it was literally in the room with them? Like they were using an app on their phone or whatever right? I dunno
The real power if chat bots lies in internal knowledge management. All of this public-facing stuff is so cringe (and unambiguously morally bad), yet the power of a chat bot being able to give you a rundown of previous projects on topic X done at you company when you have even just 200 employees can save so much time and effort. Scale this up to large corpos with 10k+ workers and it is a life changer.
Yeah this would be an absolute game changer in my industry of Life Sciences. Being able to quickly get caught up on something at the lab you missed or me the tech guy being able to quickly pull obscure information about the software I support that has been documented *somewhere* but nobody remembers where? Fuck it can't come soon enough for that.
@@kylegonewild It already exists, and its not even that expensive (20-ish €/$ per user from what I have seen). Not sure how well it works since at my workplace we only recently floated the idea and it hasn't gone anywhere yet.
It would be so amazing if the large scale management could exist without directly threat the entity of the creative industry
But unfortunately we must put all those lives on the line
Cause "line must go up"
"more up=more money"
"more money=more gooder"
I don’t like AI
‘I can’t do that Dave’
I'm not a snitch but I heard AI say it didn't like you either
Homophobe
Brave to have that written down on the internet so that our future AI overlords know to track you down and terminate you. I for one welcome our future AI overlords.
Your comment got stolen by a bot 💀
hating on AI is like hating guns, dude. its a tool. its humans that are corrupt.
I like messing with chatGPT sometimes, I'm a little disappointed I missed this. I'm not as strongly opinionative on AI as some people, I think it can be fun to mess with but I wish AI images weren't quite so easy to mass produce (anyone who can make an search tool that filters out most AI images could get a lot of use).
Also I have no idea why anyone trusts what GPT says. That's not how it works. If it 'can't lie', that's only because it has no concept of 'the truth.'
AI art must revert to its old state where it made things humans could not, just incomprehensible album cover esque art.
I do love that stuff.
Just a soup of something vaguely real. The dreamlike quality of old AI is lovely.
So, the trolls interacting with ChatGPT trained it into shit-flinging lunacy. Checks out for internet behavior
Yeah, my computer keeps trying to get me to use their “copilot” while searching the internet, and I’m sitting here like “bruh, I’m already subjecting myself to a search algorithm, why would I want to rely on a further layer of algorithms that is even less likely to show me what I am searching for and can’t even give me links to sources?”
I think what bothers me about AI the most is that for some reason, despite it being a relatively easy thing to use and whatever you want to use it for takes little to no effort. (Like face-swapping) every program I have seen is forcing you to pay so much money just to be able to use it more than once. And that's what pisses me off the most.
What am I even paying for? An algorythm? You're not manually face swapping the images I give you. You are literally just putting up a paywall to piss me off.
“Talking Spanglish without prompting” is hilarious. ChatGPT ES UNO DE NOSOTROS 😂
I for one fully intend to never deadname Vaush A Friedrich again
Isn't this the guy that wants to abuse my oshi pettan vtuber as a mighty horseman? I can't believe you've done this!
You good bro?
I tried using chatgpt to find what family guy episode a specific bit came from and it just made up fake answers and scenes in episodes that never happened despite citing real episode names and numbers lol
I think generative FILL in photoshop is/could be an excellent TOOL for making composting easier, but not replacing the artist.
I really like the LLMs. I use them to turn what I write into something people can read. It's not great and I have to sit with it for a while sometimes just to tune one paragraph. But the disability made being an author impossible. Suddenly all the stories in my head can come out. The AI tools aren't built for what I need them to do, but you can strong arm them into doing it.
It is still using a flat head screwdriver to pry something.
ChatGPT cannot handle Vaush. Someone asked ChatGPT about Vaush and it melted down.
4:30 ish, chat not understanding a deterministic system and conflating that with the idea of determinism in terms of the universe, literally gave me the brain worm from Baldurs Gate 3. Holy shit people are stupid
Wtf how didnt i hear about this???
ChatGPT keep talking like that gonna be a US presidential candidate
TLDR: AI can have a stroke
ChatGPT got into Dada poetry
The Ai bubble is probably gonna burst in the next couple of years.
Nah, it's just going to keep growing.
you know AI isnt going anywhere right, why would we go back on advancements in tech lol
I wish, but it will be what wipes us out...mark my words.
@@AsheTehFox I'm not saying it will disappear or can't be useful but the extent of its capabilities has been massively overstated and its limitations are becoming more and more apparent. It also is going to have to be heavily regulated for a variety of different reasons.
@@steve94183it’s almost like this technology is currently the WORST that it will ever be and people that think like you are in denial
The tech priest are real silent after this
Every time I hear anyone say that ChatGPT is going to get better, I can't help but feel like it won't. Like, it I suppose it might become more accurate (through a lot of error correction work by actual people) and more naturalistic (through the same process). But it's not going to be better at making quality writing, or better at making paintings.
We are going through this process the wrong way. We're trying to push art through a mathematic array. This is NOT how art is made. This is how slop is made.
Finally, since AI cannot scrape AI, it means that AI is always, ALWAYS, going to fail eventually. Model Collapse is inevitable.
gemini literally hallucinates almost all of its responses and there are no LLMs that can do anything outside of make text that sounds like a human wrote it because that is their sole purpose and it is all an LLM can do, there is no making them think or solve equations etc.: you will need another model for that. no model can be perfect though, and no LLM has been better than a human in any regard, even a very stupid human. many LLMs have little to no memory at all, including google gemini which has literally no memory whatsoever.
The exception in STEM is palaeontologists, most palaeontologists I've spoken to HATE AI 'art' and respect artists more than most STEMs due to our reliance on palaeo-art.
The entire AI learning algorithm is basically that Rick and Morty episode where Rick’s decoy family keeps copying itself.
Thanks to AI I now have to look though AI slob articles and nonsense question and answer sites on top of all the ads before I find an actual authentic source for my silly questions that I ask myself at 2 am
I said it before, I'll say it again, all AI generated content should come with some kind of watermark that can identify them as such. This would be in the interest of AI systems themselves, as it would prevent them from using previous AI content as source for their own stuff. And once this practice becomes common place as a simple measure of "AI hygiene" then we can start coming down like a ton of bricks on any "unmarked AI content" to combat misinformation.
Sir, the AI became schizzo
This is just Metal Gear Solid 2's glitching Colonel for this generation
The wiki thing is exactly why I avoid wikis for stuff I care about. Fans tend to write stuff that go infinitely more in depth anyways. I only find wikis useful for certain things like links to sources and similar stuff like it. If I want lore, I go elsewhere.
Seems like ChatGPT stumbled across the incoherent ramblings of DGG'ers...
I think AI art can be useful for dnd for making quick character art or painting a picture of the environment if you can’t easily find pictures or art online. Such generated images should probably not be posted online or be used anywhere outside the game they are being used in, however.
Having looked up stuff related to a game recently, the articles I have seen which seem to have been written by AI have little cohesion and brevity. It feels like it just gets all the info it can find about a subject and throws it in there, with no consideration of what would be most interesting to a potential reader
It just keeps saying pony, pony, pony, pony
18:23 its popping up in the publishing industry too. Tor picked up this book Gothikana (btw it's a horribly written book) and its been revealed that it used AI "art" for the cover.
Agents and so on have said they're refusing AI written work along with publishers but we all know that that will change once an AI written book becomes super popular wnd makes a whole bunch of money for it to become a trend in publishing because people don't know how to read or think critically about books.
12:40 predator comes back to the crime scene
HorseGPT should be the next best thing after this.
😂😂😂
I appreciate the clarification in the end. Most of the time he doesn't do that, just makes some statement about STEM in general. I'm in STEM and very much appreciate creative types including artists. I create tools and they use those tools to make something beautiful that makes life in this cold universe occasionally tolerable. As such, I agree with Vaush that when the tool becomes a replacement for the artist, or a replacement for human expression, it has usurped its purpose.
AI said the first ever Beyblade release was Driger S. Completely wrong. It was ultimate spin dragoon. When corrected the AI apologised and simply repeats what you told it
There are multiple lawsuits going right now. There are also a bunch of news articles claiming it was dismissed, but it was actually a statement that there were issues that needed amending and to come back with an update. The core of the cases are still going through just fine.
Also there's the go fund me still running