The AI generator museum of uncanny amusement
-
LivDiv wrote:Getting people to pay for this stuff could be a road block. Coding eggheads are about to come up against what creatives have for years "do it for exposure", "my son knows Photoshop", a general perception that stuff that is fun to consume is fun to make so is it's own reward. It's probably in the Limewire stage right now, jank, copyright infringing minefield. Will need to be iTunesed or Spotified.
I think you're right on getting people to pay, and see a future where venture capital fuelled AI services struggle to get paying customers to cover the bills. And so, they do what all those social media platforms did when they had users but no cashflow - sell to advertisers.
Imagine it, free AI image generation but every image contains branding or product placement from an advertiser.
A man types Princess Leah in a gold bikini and 2B holding hands in the style of Frida Kahlo and he waits. An image is rendered. He tweaks it, runs through various outputs until he finds the one that is just right.
He clicks to save it. He opens the file. It's the picture he wanted, but now in her free hand Princess Leah is holding a cool and refreshing Bud Light.
Everything is good for a while. Users flock. The only limit is your imagination. But then a news report. It turns out AI can be used to make sexualised images?!?!?
Advertisers threaten to flee. The companies shackle their AIs. Images of violence? Fine. Barely legally distinct Disney mascots? Sure. But nipples? Good god no. The platform becomes gentrified and "safe" even your dad starts using it.
There are more free and less restricted services that pop up, without all the ads and censorship maaaaaan, but they run like dogshit and always close down or sell out once it is clear that people will absolutely not pay.
The number of ads increase. Before the release of her latest album, all women in AI images are changed into Beyonce. The Guardian write approximately 4972 stories about how brilliant this is.
The ads grow again. Suddenly you're getting 2 Suggested Pictures between every prompt.
Because you want Man in cowboy hat holding laser gun and riding unicorn that is rearing back. Rainbow in background we thought you might like Images of you driving in the brand new BMW X3 along a road made of Peanut M&Ms.
A chinese company comes out with a new AI image generator that lets you take your AI image generated content and add it to selfies in real time. Every teenager in the world downloads it.
America moves to ban it because it could be communist. Every time the company is mentioned online, 87636383 men reply with edgy comments about social credit scores. Israel calls it anti-Semitic because young activists use it to insert Palestinian casualties of war into anti-war marches, so that they can be there in spirit and remembered.
The boycott clears congress and the senate. Something about data privacy and national security. None of the congress people know exactly what data or are able to point to any evidence that it is being harvested but, you know, communism.
Meanwhile ChatGPT, now owned by the Facebook-X-Warner multimedia conglomerate is found guilty of illegally surveilling 350 million American voters during an election year and inserting disproven conspiracy theory symbols into their image outputs without disclosing it. They receive a 450 million dollar fine and promise to take a long look at their internal processes so that it never happens again. It barely makes the news, though, and 3 years later Facebook-X-Warner are fined another 400 million when it is discovered they commissioned millions of fake AI news stories over the time period surrounding the original fine in order to bury any actual reporting on the case. They promise to take a long look at their internal processes so that it never happens again.
Everything settles down. The services, ad-filled and barely useable remain pumping out endless reams of content. Any time a new movie, book or game masterpiece is made, a thousand AI legally distinct imitations are created based directly on the original source and sold at a lower price.
Eventually it all becomes AI. An AI Ouroboros. AI movies being reviewed by AI writers and posted on AI created websites, posted onto AI controlled account on AI moderated social media platforms. Ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads within ads -
Repeat after me.
Interlinked -
Like social media but then the machine becomes 'conscious' and tries to wipe out humanity. We don't understand what's happening and keep debating the 'how' and 'why'. Keeping the AI's running is profitable so why not as they run all of our integral infrastructure not to mention our social media. Plus we luv ads, and cheap stuff.Steam: Ruffnekk
Windows Live: mr of unlocking
Fightcade2: mrofunlocking -
Keeping the AIs running is absolutely not profitable, unless you're Nvidia and selling shovels/DGX units.
-
Can't find the link now but there was a report that brands don't want AI and are stipulating no AI use in their contracts with ad firms. Main reasons were not wanting to encroach on some one else's IP, not wanting their IP fed back into AI models and becoming easier to reproduce. And the negative PR that comes from AI use at the moment.
-
@djchump
But you won't have to pay any workers! Think of the glorious profits!
There won't be any consumers either so they'll try to save capitalism with a ubi scheme and free trade wall street esque market where all of the ubi plebs can participate and 'fend for themselves' for riches and glory.Steam: Ruffnekk
Windows Live: mr of unlocking
Fightcade2: mrofunlocking -
Yeah when things calm down a bit it'll be the big tech players with the mature models that can be used for pro-tier work. And they'll slow the R&D down to keep costs manageable.b0r1s wrote:Ouch! Sounds like incredibly bad management. As the compute needs accelerate there’ll only be space for the big boys.
It's a bit like the move to streaming but accelerated. Chuck money at it, try and get market share and force the competition out, realise there's not as much money in it as you thought, flounder a bit, cut costs.
The next few years of this will be a garbage fire.
-
@Monkey
If you remember the link id be keen to read that but I can well believe it.
Id also argue its not accurate enough for the microdetail brands require. -
Got it. Buried in my twitter likes.
https://adage.com/article/digital-marketing-ad-tech-news/ai-restrictions-added-ad-agency-contracts/2548696 -
-
That's weird. It's not paywalled from Twitter but is from the direct link.
-
No biggy.
This made me chuckle.
“To suggest that the agency should disclose and obtain the advertiser’s approval for any such use serves only to unnecessarily complicate the contractual requirements when there is no practical benefit for the advertiser in doing so,” Karandikar said
Nothing new there then. -
monkey wrote:That's weird. It's not paywalled from Twitter but is from the direct link.
Quite a few places do that. Means people can read something being shared, and might end up subbing. -
monkey wrote:Can't find the link now but there was a report that brands don't want AI and are stipulating no AI use in their contracts with ad firms. Main reasons were not wanting to encroach on some one else's IP, not wanting their IP fed back into AI models and becoming easier to reproduce. And the negative PR that comes from AI use at the moment.
That's not true though, a lot of our clients are happy to use AI, and our job is to educate them on how and the ethics of it. The article is correct in that it needs to be clear in contracts, but usage of it, which does save money, is definitely happening across the industry. -
AI is like any computer tech, stupid expensive at first and cheap as chips later in the same way that room sized computers in the 70's gave way to desktops and eventually mobile phones. It can't be controlled when it's in the hands of everyone.
In the meantime it's parallel nature means people can collectively train a complex model. This is already happening with individual GPU users on places like Reddit contributing to a single model.
What has happened is generative has exploded so quickly because big tech has thrown massive amounts of resources at it just to see what it can do. It'll be in the hands of everyone soon as the price comes down and we won't have to rely on big tech to provide it.
The main thing is that AI is very simple to do technically and it's available to everyone already. It's just that big tech has the newest chips and generative AI has artificially pushed the price up, for now. The cost of training will also come down as the current models are extremely inefficient. They just chucked a shitload of parameters at it to see what it could do for a lark, not even sure that it would work."Plus he wore shorts like a total cunt" - Bob -
It really is so early days."Plus he wore shorts like a total cunt" - Bob
-
They don't even know how it works, not actually. The models are too big for humans to understand.
-
Is anyone doing a "Train it yourself" type product?
This is what will appeal to big business ultimately. For example Nike might want to train their AI purely on Nike products so "A red sneaker being held by Post Malone" would only come up with a shoe that looked like a Nike shoe.
Or DC may want an AI that can be told "an ensemble of heroes fight a planet sized villain" without it pulling in Marvel characters or even art styles that aren't theirs.
I realise better prompting helps, "sneaker" could be "Air Max 1" but a clean data set would be potentially more efficient and reliable. Would alleviate some of those issues around things being kept in-house as well, Nike and DC don't want to feed their archives into something anyone can access. -
b0r1s wrote:They don't even know how it works, not actually. The models are too big for humans to understand.
They know how it works, it's just differentiation. They just don't know what it's doing."Plus he wore shorts like a total cunt" - Bob -
LivDiv wrote:Is anyone doing a "Train it yourself" type product?
This is what will appeal to big business ultimately. For example Nike might want to train their AI purely on Nike products so "A red sneaker being held by Post Malone" would only come up with a shoe that looked like a Nike shoe.
Or DC may want an AI that can be told "an ensemble of heroes fight a planet sized villain" without it pulling in Marvel characters or even art styles that aren't theirs.
I realise better prompting helps, "sneaker" could be "Air Max 1" but a clean data set would be potentially more efficient and reliable. Would alleviate some of those issues around things being kept in-house as well, Nike and DC don't want to feed their archives into something anyone can access.
Big non-tech firms will catch up soon enough and use their own models from their own datasets. You can actually tweak an existing model as it is but it'll take a little time. I downloaded a Hellman's app that uses your phone camera to scan the contents of your fridge and cupboard and come up with suitable recipes that use mayonnaise.
We'll start seeing this kind of thing all over the place and all companies big and small will start using their own custom models. At the mo though big tech is grabbing all the chips. I'll be interested to see what the gtx 5000 series can do."Plus he wore shorts like a total cunt" - Bob -
What are they training it on? I thought big tech was gobbling up and locking away all the big data sets. Is it 'publically available' (aka stolen) data?SpaceGazelle wrote:AI is like any computer tech, stupid expensive at first and cheap as chips later in the same way that room sized computers in the 70's gave way to desktops and eventually mobile phones. It can't be controlled when it's in the hands of everyone. In the meantime it's parallel nature means people can collectively train a complex model. This is already happening with individual GPU users on places like Reddit contributing to a single model. What has happened is generative has exploded so quickly because big tech has thrown massive amounts of resources at it just to see what it can do. It'll be in the hands of everyone soon as the price comes down and we won't have to rely on big tech to provide it. The main thing is that AI is very simple to do technically and it's available to everyone already. It's just that big tech has the newest chips and generative AI has artificially pushed the price up, for now. The cost of training will also come down as the current models are extremely inefficient. They just chucked a shitload of parameters at it to see what it could do for a lark, not even sure that it would work. -
SpaceGazelle wrote:b0r1s wrote:They don't even know how it works, not actually. The models are too big for humans to understand.
They know how it works, it's just differentiation. They just don't know what it's doing.
That’s what I meant. They don’t know why the LLMs make the decisions they do. -
monkey wrote:
What are they training it on? I thought big tech was gobbling up and locking away all the big data sets. Is it 'publically available' (aka stolen) data?SpaceGazelle wrote:AI is like any computer tech, stupid expensive at first and cheap as chips later in the same way that room sized computers in the 70's gave way to desktops and eventually mobile phones. It can't be controlled when it's in the hands of everyone. In the meantime it's parallel nature means people can collectively train a complex model. This is already happening with individual GPU users on places like Reddit contributing to a single model. What has happened is generative has exploded so quickly because big tech has thrown massive amounts of resources at it just to see what it can do. It'll be in the hands of everyone soon as the price comes down and we won't have to rely on big tech to provide it. The main thing is that AI is very simple to do technically and it's available to everyone already. It's just that big tech has the newest chips and generative AI has artificially pushed the price up, for now. The cost of training will also come down as the current models are extremely inefficient. They just chucked a shitload of parameters at it to see what it could do for a lark, not even sure that it would work.
It's not stolen, no. There are a bazillion datasets available on the internet. The LLMs are literally trained on the same internet pages available to everyone.
People are collaborating on various models. Some are using astronomical data, some financial, some climate change. Pretty much anything really."Plus he wore shorts like a total cunt" - Bob -
This forum probably has enough pages to imitate a forum member. You could do it with a mix of all their posts and a little GPT."Plus he wore shorts like a total cunt" - Bob
-
Yeah but you've got to pay Gman tuppence per post scraped.
-
What big tech has is information on you personally. This can be very powerful and it could get vast amounts of data on you from your say, your phone. And it'll know your relationship with other people extremely well."Plus he wore shorts like a total cunt" - Bob
Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Categories
- All Discussions2,718
- Games1,881
- Off topic837