Mind blowind OpenAI's text to video SORA.

[tom66]

That's not realistic, given that the silicon needed to train these models takes up entire warehouses and didn't exist only a few years ago.

What I'm saying is that they might have better algorithms or custom hardware.

It's been a long time since the three-letter agencies had access to better tech than the public.

If you're in a position to know you shouldn't be revealing such secrets!

It has been said by someone close to OpenAI that 30 seconds of SORA video currently costs around $500-700 in cloud compute resource, therefore it's likely to remain a closed beta for some time before they can figure out how to lower the cost and scale it to the massives.  SORA appears to be a well-converging DALL-E 3 (it shows similar artefacts to DALL-E 3, suggesting common datasets).

As for access to specialised hardware: yup Nvidia make server-specialist cards just for AI.  These are certainly out of the reach of the masses, because they have to be sold as part of a whole system, which probably costs nigh-on $50k per server.

[Neutrion]

I again ask myself whether the pros and cons of this tech. are in  balance. And I don't see it.
Pros: It will be cheaper to make movies, and ads. Some new artform will also develop itself, which will be completely dependent on these thech giants, censorship, payment, etc.

Cons: People within 1-2 years will not belive anything which they don't experience themself. Basically everything what broadcasting technology brought to us will be lost on the individual level.
That will mean complete social breakdown and chaos. (Just think about the flat Earth believer at the moment.)

I don't think people creating these technologies even think about the consequences what they do. They just do it because they can. Or the are afraid of "the enemy".
And  all the features are aiming to replacing ANY need for humans. At the end the developers as well. Yes there is work going on  modells which can replace even mathematicians, and I suppose also electrical engineers and the rest.
Some controlling will be needed at the beginning but the human capabilities of creating and understanding anyhing without these tech giants will be degrading rapidly. (Like when people completely forgetting map reading using the navi.) Possibly even supressed, as those not controlled fully will be suspicious and possible threats.



Even if no superintelligence would be created,(which is also a real danger) everything will be controlled by very few people on the level not imaginable even in dictatorships. I think if we don't stop this, our species will be extinct or living under nightmarish conditions very soon.

[KE5FX]

Totalitarian dictatorship may or may not result from widespread development and deployment of this particular technology, but it will absolutely be required if you want to stop it. 

What I don't get is why so many people actually want to do a robot's job... or a dictator's.   

[tom66]

It's certainly going to be an interesting time.  I don't know how exactly to feel about these models.

On the one hand, it might well be possible to generate any art you could conceive of... for instance, "Make a version of Iron Man where Mr Bean plays the role of Tony Stark and accidentally creates the armoured suit whilst fiddling with his toothbrush."    The options for creativity are huge.   It will allow the output of an animation studio to increase several fold.

But on the other hand it's easy to see how this will decimate many industries, because it's arguable that there is only so much content that people can consume (otherwise it just becomes noise), and AI threatens huge parts of the creative community here.  Films like Iron Man have the budget they do in part because thousands of artists work on the animation process.  Feasibly they might be replaced by 10 people training and tweaking an AI model.  It's a potential bloodbath.

I am seeing some commentary from artists around AI training being banned on their copyrighted work, but this isn't currently illegal.  This might be a case that makes it to the Supreme Court in the US in the near future (and other jurisdictions,  Getty are suing Stable Diffusion creators StabilityAI).  But the cat is out of the bag on this.  You can train an SD model at home on a <$1,000 graphics card, and generate imagery in seconds.  So even if large-scale training by companies in the UK and US is banned, this technology will just move elsewhere.  And Hollywood insiders will complain that they can't compete with Beijing or Mumbai's new film outputs, and legislators and unions will get beaten down into allowing the technology eventually.  It does feel unstoppable.

[MT]

Totalitarian dictatorship may or may not result from widespread development and deployment of this particular technology, but it will absolutely be required if you want to stop it. 

What I don't get is why so many people actually want to do a robot's job... or a dictator's.   

Pick a future "war theater" then press "False Flag videos" and incoming in 5-4-3-2-1.

[KE5FX]

The effect will be to teach people that they can't blindly believe anything they see on a screen or hear over a speaker.

I see that as a positive effect.

Relax.  This is what Joe Strummer J. L. Borges trained us for.

[Bud]

Train train train, the keyword here is "train". What you going to train your stupid AI on if there will be no human generated material to use for training.

[tom66]

Train train train, the keyword here is "train". What you going to train your stupid AI on if there will be no human generated material to use for training.

Zero-shot learning; GPT has already demonstrated it can train off its own data.

[Neutrion]

Totalitarian dictatorship may or may not result from widespread development and deployment of this particular technology, but it will absolutely be required if you want to stop it. 

What I don't get is why so many people actually want to do a robot's job... or a dictator's.   

Can you define a human job under these circumstances, and in the near future?

The effect will be to teach people that they can't blindly believe anything they see on a screen or hear over a speaker.

I see that as a positive effect.

Relax.  This is what Joe Strummer J. L. Borges trained us for.

So you regularly watch your presidents/governors/supreme courts speeches in real life? Because that is what you will  need, that is what I am talking about. Or there will be truth agencies which tell you what is of what you can see the truth and what is not. Not good in a democracy.

Even if there will be some methods to find out what has happened and what not, you basically have to treat everything what you see as possible fake. So the basic human judgement will loose its value completely. This is what society is built on. But hey why bother, if one can make video production easier?

Train train train, the keyword here is "train". What you going to train your stupid AI on if there will be no human generated material to use for training.

Zero-shot learning; GPT has already demonstrated it can train off its own data.

And using big data on multiple dimensions it is not a big deal to filter out humans. Mobile phone total surveillance is also good way to do it. Every connected data basically, from every sensor.

[Muxr]

Train train train, the keyword here is "train". What you going to train your stupid AI on if there will be no human generated material to use for training.

They are already doing this. They are making synthetic training data, generated by AI.

[berke]

And using big data on multiple dimensions it is not a big deal to filter out humans. Mobile phone total surveillance is also good way to do it. Every connected data basically, from every sensor.

I have an idea, I know it will sound a bit extreme but what if there was a law forcing people to stay and work from home?  Even if it was for just a few months.  Imagine the goldmine of data (voice and video) you could get from all those Teams/Zoom calls.  There could be an exemption for people whose jobs don't produce trainable data anyway.

[BrianHG]

[RJSV]

   'Old scanned in comic books...'. - Brianhg

   God help us, if AI ever gets hold of old, 1970's
'ZAP COMICS' copies....heh heh heh
   ("You can't pick your friends nose", and other, Ron Crumb topics, too graphic to discuss here)!

[Bud]

All this is good for is another Shrek movie, that is it.

[BrianHG]

Finally, some people get to use SORA.  Here are some of the results...

[SiliconWizard]

Train train train, the keyword here is "train". What you going to train your stupid AI on if there will be no human generated material to use for training.

They are already doing this. They are making synthetic training data, generated by AI.

Oh yeah. Some kind of digital consanguinity. Sounds good.

[xrunner]

Oh man! If you're a political dirty trickster you GOT to get access to one of these systems! 

[jonovid]

still need human artist's to do cleanup!
some technical details that need fixing to make the videos more believable. 

[BrianHG]

[soldar]

AI is developing in various fields, not only graphics creation. Also in language creation and translation. Also in driving cars. Also in robots of the two legged and four legged types built by Boston Dynamics, Honda, etc. I just saw some YouTube videos of robots made by a Chinese company, Limx Dynamics, and they are quite impressive.