I'm kind of excited to see how scifi authors will tackle the generative AI revolution in their novels.
As of now, the models still need large amounts of human produced creative works for training. So you can imagine a story set in a world where large swathes of humanity are regulated to being basically gig workers for some quadrillion dollar AI megacorp where they sit around and wait to be prompted by the AI. "Draw a purple cat with pink stripes and a top hat" and then millions of freelance artists around the world start drawing a stupid picture of a cat because the model determined that it had insufficient training data to produce high quality results for the given prompt. And that's how everyone lives their lives....just working to feed the model but everything consumed is generated by the model. It's rather dystopian.
I have a novel I've been working on intermittently since the late 2000s, the central conflict of which grew to be about labor in an era of its devaluation. The big reveal was always going to be the opposite of Gibson's Mona Lisa Overdrive, that rather than something human-like turning out to be AI, society's AI infrastructure turns out to depend on mostly human "compute" (harvested in a surreptitious way I thought was clever).
I've been trying to figure out how to retool the story to fit a timeline where ubiquitous AI that can write poems and paint pictures predates ubiquitous self-driving cars.
I would say it's very profitable in terms of ideas...if you put the work. The problem is that most main-market sci-fi is not about ideas, but about cool special effects and good vs bad guys.
> As of now, the models still need large amounts of human produced creative works for training.
That will likely always be the case. Even 100% synthetic data has to come from somewhere. Great synopsis! Working for hire to feed a machine that regurgitates variations of the missing data sounds dystopian. But here we are, almost there.
Agreed, by some definitions, specifically associating unrelated things, models are already creative.
Hallucinations are highly creative as well. But unless the technology changes, large language models will need human-made training substrate data for a long time to operate.
It's bimodal. AI can automate a lot of low level knowledge work, but as wide and deep as its knowledge is, it is also incredibly superficial when it comes to logic and creativity. What it's going to do is hollow out the middle class, as creative people who know how to wield AI will become wealthy while the majority of white collar workers are forced into trades.
A major follow up to GPT-4 later this year is rumored to be (far) superior at logical reasoning than GPT-4. What's likely to happen if that becomes real?
That might let it encroach more into some fields like law where it's almost good enough already. Shitty time to be a junior lawyer, firms are going to hire and promote people not for their legal skills but for their ability to manage/attract clients.
In general though, I don't think the extra reasoning ability is going to enable it to displace that much farther than it already will, GPT lives in a box and responds to prompts. When it's connected to multiple layers of real-time sensor data and self-directing, that'll be another story.
There were independent efforts to create AI agents since last year as well. AutoGPT and BabyAGI iirc. They didn’t go far probably because the LLM used was not good enough for that.
> AI is automating the creative, intellectual work and leaving the rest to us.
Indeed, there is a risk it completely devalues creative jobs. That's ironic. Even if you can still use AI creatively, it removes the pleasure of creating. Prompting feels like filling Excel sheets while also feeding a pachinko machine.
if it was actually AI, instead of a stochastic parrot, we could ask it to design robots that could do the manual labor that we still have to do, because we haven't been able to design robots to do the manual labor.
Unfortunately, LLMs aren't intelligent in any way, so you cannot ask them to synthesize any kind of second-order knowledge.
This is why they won't take away the creative work, either. They are fundamentally incapable of creating anything new.
This is the beginning of the end for many of them too. Look at the opening line of the page:
> We’re teaching AI to understand and simulate the physical world in motion, with the goal of training models that help people solve problems that require real-world interaction.
Text-to-video is just the flashy demo that everyone can understand after exposure to text-to-image. Once the model can "simulate the physical world in motion" it's only a few steps away from generic robotic control software that can automate a ton of processes that were impossible before.
Humans still have the benefit of dexterity and precise muscle control but in the vast majority of cases robots can overcome those limitations with better control software and specialized robotic end effectors. This won't soon replace someone crawling under a house or welding in awkward positions, but it could for example replace someone who flips burgers or does manual labwork.
This could eliminate the limiting factor for automating many manual processes. (ruh-roh)
Less glibly, no matter how good you are at following instructions, tearing out a wall filled with water than can destroy your home, fiberglass insulation that can damage your lungs and electrical wiring that can kill you will never be something I’d recommend a layman do. No matter how good the ai tutorials are.
Not when intelligence is cheap and highly abundant. Perfecting general robotics as an improvement on humans will be quick. The upper limit of strength and consistency is much higher.
Today in the real world AI can replace very little of designers, programmers, etc. Lots of potential and extrapolation, sure. but hasn't happened. What has actually been produced by AI has been panned as not quite ripe yet.
Same with robotics. Lots of potential, but hasn't happened yet. If you read the description, Sora, is based out of trying to simulate the physical world to solve physics based problems. Something that would be perfect for the next leap in robotics.
Think about it. Sora demonstrate AI can understand real world physics to a scary degree.
If you use Sora like models to imagine what actions needed to be taken, then realize it, well, the only thing left is to create an arm/fingers that can took action, then you are done.
Machines have replaced a lot of blue-collar jobs alright. It's just that most of it happened during the Industrial Revolution, so we aren't even aware of all the shitty (and not-shitty-but-obsolete-nonetheless) jobs that used to exist.
It's automating some of the craftsmanship part, which is substantial, but in a sense, it also threatens the creative part.
It's already very tempting for large entertainment businesses to create lazy remakes as it involves less risk. Automating creative jobs will create a shift at the production level but also on the receiving end: the public.
that would never happen because someone owns the robots and rich people can afford more robots than poor people and rich people aren’t rich people if poor people aren’t poor
Come on, don't you see that the capability to understand the physical world that sora demonstrates is exactly what we need to develop those household robots? All these genAI products are just toys because they are technology demonstrators. They're all steps in the way to AGI and androids.
sensorimotor control is imo not at all the bottleneck. Teleoperated androids could do lots of useful things right now, but the AI is lacking to automate them.
well I let's say you want to make the coffee and we split that task into roughly two subtasks. The first is to imagine what motions are necessary to do that. How does the coffee cup have to move, how does your hand have to move to grasp it, etc. The second part is to find a way to use your muscles or actuators to execute those imagined actions.
I claim that the first part is the more difficult one and where we have the bottleneck currently. Furthermore, generative video AI is exactly the kind of thing that would give a model an understanding of what kinds of things have to happen in order to make coffee.
Somehow, according to that logic, and in general the logic of all AI danger hysteria, humans have no agency in determining what the limits of what AI is fed and of its use and abuse.
Some humans do - the investors and executives in AI tech companies (and the legislators who theoretically could regulate them) , who all stand to make a lot of money from every one of the "AI danger hysteria" scenarios, and are therefore highly motivated to bring them to fruition.
The rest of us have no choice. Despite millions of artists, animators, etc. all being resoundly opposed to AI art, the models that infringe their work are still allowed to exist, and it seems they're fighting a losing battle.
A lot of people are being "hysterical" because a lot of people don't have a choice.
To be clear, the problem of these scenarios is tightly intertwined with the problem of unfettered capitalism and wealth inequality in general. Food and shelter require money, and we get money by working a job. If millions of jobs disappear overnight, then of course millions of people are going to be distressed over no longer having ready access to food and shelter.
The idea of "just getting another job" doesn't scale to the destruction of entire industries employing tens of millions of people. This is how depressions are made.
The idea of "the depression will end someday" is not only not necessarily true as wealth inequality skyrockets, but is also cold comfort to the people who will lose their houses and for some, lives, due to the disruption.
A different economic system could perhaps allow us to appreciate these technological advances without worrying about them displacing our ability to live. But the American political system consistently and firmly rejects any ideas not rooted in social darwinist capitalism.
For your sake, I hope your resume is very impressive.
If millions of jobs disappear overnight it means AI is amazingly good, which means people will also have AI empowerment on a whole new level as open source trails companies by 1-2 years. Everyone will just order their AI "take care of my needs", maybe work along with it. You got to agree that we already have some amazing open models and they are only getting better - that empowerment will remain with us in times of need.
"Companies employing people" will be replaced by "people employing AI". Open models are free, small, fast, trainable and easy to use. They capture 90% of the value at 10% the cost, and are private.
"Companies employing people" getting replaced by anything is pretty dangerous in an economic system where employment is synonymous with having food and shelter. It won't matter that AI could help me keep a to-do list or generate pretty videos if I don't have a job or income.
What we're looking at is a massive decrease in the relative economic value of the average human's work. If the economic value of a hundred people is less than what the company can produce with a single human operator running AI models, then those 100 people are economically worthless, and don't get to eat.
We drastically need to tax the usage of AI models on the huge windfall they're about to create for their operators, and use that to fund universal basic income for those displaced. Generally speaking, as automation and wealth disparity skyrocket, UBI will be required to maintain any semblance of the society we currently have. I am incredibly pessimistic about the chances of that happening in any real way though.
I would agree. While we are seeing all this creative work get automated by AI, how big of an impact would that really have on the economy?
Fully-functional autonomous driving will have a much larger economic impact - and that's just the first area where autonomous robots will come into our lives.
But in reality it seems like the opposite is going to be true. AI is automating the creative, intellectual work and leaving the rest to us.