Yep. We will have Degrees and will have nothing... p.s. Someone who programmed ANN in VHDL long time ago (I was young, I didn't know!)

This is why experts keep warning that AI has enormous potential to slip into uncontrolled chaos. Because algorithms can evolve exponentially in terms of complexity, there is a real possibility of this complexity to become overwhelming to society. At that point, it will need to be so highly regulated that it becomes almost useless. That’s the reason why so much warning. Experts don’t want this to happen. Sadly, politicians aren’t paying attention and industry is left to self regulate. You can see this best right now at google’s Gemini LLM. It is exceptionally restrictive at the expense of being competitive with Microsoft’s Bing.

Which is why they need the support of the open-source community. The alternative is that we are locked into a Microsoft-OpenAI walled garden which will control what you get and how much you must pay for it, plus the added cost of total access to your data and the results of your queries.

how can we donate?

Facts

​@@philsomething8313 subscribe to them to get early access and commercial access to new models

They even inherited their censoring!

What a great time to be alive!

🐂💩

I came to make the same comment. Bro needs to put a warning next time 😮😂

They confirmed it will "SFW" unfortunately. So it will be up to the community to put the N in the NSFW

The problem is that there isn't one truth or one answer to every problem. Humans design with a purpose but AI emulate us without understanding fully. We disagree about things. Change our minds with time and trends. AI + human > AI.

@@hombacom There is nothing magical about human cognition. AI in time will be able to surpass us everywhere, since humans are hardware limited.

Given how most people misunderstand why Einstein got a Nobel prize, I wonder if AIs answer that one correctly (given that they use databases built off of the internet)

I find capability tends to scale somewhat logarithmically with model size, so brute force scale is not terribly efficient on it's own. Still very impactful off course, but there's a reason why models size tends to grow explosively, making everything awfully heavy to run. I'm going to seriously miss the old days of casually cranking out 9+ images for a quick test run every time I tweaked the prompt in SD1.5. That little baby is comically lightweight by modern standards.

​@@fnorgen Yup, and that's why I am still using mostly SD v1.5. I can still render 8 images at once in SDXL, but it's not as fast. SDXL also doesn't like long prompts when I want to be very specific. From one of the examples in this video, I saw a very long prompt and the results matched quite well, which gives me hope. 🙂

the big jump from SD1.5/2 to SDXL was that you could use more natural language to describe your image. SD3 seems to improve on that alot

@@AngryApple I hope so! That would reduce the trial & error, and make the tool more productive. ^____^

It's a long repeated letter, to be honest a small error

I mean, you will have the option to disable T5 and CLIP-G/14, leaving you with just CLIP-L/14 like SD1.5, but even then it will still have far better text generation due to the higher model capacity.

@@miriamkapeller6754 i’m already nostalgic for it!

Hell, it‘s even a jump from XL!

There was a research where they put some people to watch videos and then close their eyes and reimagined those videos again on their minds while an AI was reading their brain activity, they trained the AI on that data and it was able to rebuild the "videos" the people where thinking on... the videos the AI generated weren't perfect, but they do looked like the videos those volunteers have watched before.

@@MCA0090 Wow that was amazing

Go to therapy, people. You are not in a state of good mental health.

It's not without artists. It just requires less manual skills.

@@vidyagaems4063 I mean, if you want to call yourself an artist if you're typing in a few words until it looks right, then you can do that, just don't expect anyone to take you seriously...

​@@miriamkapeller6754 You would call a mindless drone that can only reproduce reality an artist. I would call it a "camera", because actual art requires creativity - something you lack.

​@@vidyagaems4063 There's not a single word in your post that makes sense. Try again tomorrow when you're sober....

@@miriamkapeller6754people who understand creativity and art will take them seriously. Who cares about the people who have yet to catch up?

when?

​@@jonorgames6596 when it's available can't you read?

You can look at the paper and the showcases. SD3 is in a completely different league in both quality and prompt understanding, there's barely any point in even trying to compare the two.

I tried with ChatGPT to no avail

Well, not really. There's a good piece of info in the video at 5:24 to show that. See how the lines minimize and then start to go back up? You can think of that point as the point where the AI fails to model human creativity anymore, sorta. Humans are capable of observing that they're getting close to a ground truth and that their model of the situation needs additional scope or refinement to improve. Every single AI will in every situation develop its model, get to that minimized point, and then keep on going with the same assumptions even after the model fails. It's called the bias-variance tradeoff. Nobody can fix it except with human intervention. If you let a machine run on its own, it will always, no matter what, reach the point where additional effort makes things worse. Human creativity is exactly the thing that we can apply to avoid that situation. A better claim would be that the function demonstrates that a lot of what we thought was creativity was really just artists yanking on established norms or being random. But real creativity is still out there as an ingredient in real art and still wins every time.

Perhaps the entire universe is just a fractal function with astronomically large amounts of inputs over time!

What's scary about it? I seen only net positive

😍🤣😅

Dude, not cool. My wife literally left me 3 weeks ago to explore Hungary with some “friends” on a short trip. I thought it’s no big deal. But later on I found out that it was just a lie. She actually ended up going to Hungary alone with the sole purpose of getting to meet and get to know Karoly. I found this out from a letter she left me right before she left about how Karoly is the “love of her life” and how she wants to start a future with him. Fast forward to today and she has pictures of them two together on her social media and she has completely forgotten about me. I’ve never been more devastated in my life. If I would have never watched two minute papers in the first place I could have avoided this entire mess.

@@dna2dna275 what?

@@jknvorneb My wife left me for Karoly Zhona Faher. that’s “What”. My life has been ruined because this guy’s videos about AI and machine learning somehow seduced my wife.

@@dna2dna275 how did it seduce your wife

Yesterday

New version, yes. Very much improved as well.

Yeah, that‘s why there‘s a 3 in the name

stability open sources everything, youre complaining in the wrong place

so where can I find this open source Ai you mention?

Literally google it

@@gawkthimm6030 What are you asking for exactly? SD1.5, SDXL and Stable Cascade are all freely available, research paper, code and model weights. Same for all other models they put out.

There's this thing called "Google", I don't know if you use it. And yeah, Stable Diffusion is completely open-source.​@@gawkthimm6030

You can... they already announced the smallest version will have 800M parameters, so you can run that and probably the next largest one. too Not the 8B model, of course.

Not sure if it can count, but it’s supposedly far better at accurately generating complex descriptions such a “A blue monkey doing a handstand on top of a red pig next to a miniature rhino.” I would imagine the improved comprehension affects counting to some degree.

"squirrels in snow" must be an euphemism for a sex practice. Or paraphrases one. Or paraphrases one which is already paraphrased and with enough attempts end up showing exactly what perverts want it to show.

See his previous video about this same model, but basically yes, it's a little better at it

Image creater has the goofiest censorship. I can‘t seem to generate anything remotely fun.

" A lot of you will be able to run this from their laptop", lol sure :D

only if you define a human's worth by what they can produce

@@patrickhector Good point. I am not define like that, but not I am who decide

@@YYYValentine you underestimate the importance of self-worth

Only youtubers can't be obsolete because people never gonna stop watching videos doesn't matter what the world is going through

It just gives people who are creative a massive tool to create whatever they imagine. The computer won't sit there and decide what to make.

You won't believe it, but there is one actually. Search for "TU Wien Rendering #1 - Introduction". Right at the beginning he teaches his students how to pronounce his name.

Watch his ray tracing course on his channel, lecture 1 he explains how :p

Just say Dr. Károly Zsolnai-Fehér it‘s that easy

no it needs to follow prompts more accurately

@ryzikx would this include "apple on the left side of screen" or do you mean easier to prompt with natural language

It's not being out of touch 😂 AI is evolving super fast and unless you are terminally online following all the shit that's happening, you just don't know

Stable diffusion has been freely available for over a year. You don’t need to be terminally online to discover this. The only people who don’t know are people who don’t care to know.

We’re headed in that direction with Stability and others now refusing to allow free commercial use in their “open source” releases.

some are. they use SD on the backend. hell midjourney is a subscription

That is the plan long term. After all, somebody has to make money with it. Else there is no point in doing it.

They don't. That is for Midjourney, this is for Stable Diffusion.

@@zdanee Dude the link provided to subscribe for the waitlist literally has Discord Id as a mandatory field.

There is stable horde.

After watching the video: many people will be able to just run it on a laptop, and soon in phones.

Replicate might add it but it's not available yet

Just tried that on the free bing create and the result was flawless.

What are you talking about? There haven't been so many exciting new techniques in 3d in a long time. There is Nanite, Lumen, upsampling, frame generation, path tracing, realtime denoising and now all the AI stuff on top.

@@c0d3_m0nk3y- 3d has been evolving quickly, no doubt. Especially the webgpu stuff, and all the augmented reality and all that. Super fast evolution. AI outpaced it...

@@c0d3_m0nk3y Why would anyone learn how to design a 3D model now that AI has killed the need for it? This is going to be every art field. Pretty bleak.

@@dshcfh Oh, you mean from an art perspective? I thought @jeffg4686 was talking about engineering...

Can can like the dance?, a novel form of entertainment these days

Good luck waiting for someone to built your utopia. You might be sacrificed next for someone else's personal idea of their utopia - without you in it. :/

how does replacing artists with ai help humanity

Nice, the topic aside, it's good to be flexible to evidence and not just have a one way of thinking about something, everything is nuanced and having a one track mind is just a coping method to a complex world rather than an accurate way of assessing situations (Ironically we seek generality in these systems, but we struggle to have it.) AI is clearly a helpful tool, more accessible than any other, and will help many people, and already people with injuries/disabilities or lack of time or simply not wanting to spend effort on something that doesn't need it depending on their use case, do use some sort of AI every day. There are many risks to consider with almost anything too, and we should always critique new technologies even more so, but as the technology has developed thus far, the benefit certainly outweighs the downsides for many; in every area I can look, AI is providing value, such as in the gaming world with upscaling technologies (2x FPS with basically no downsides is no joke), or art and architecture being given a much faster creativity process, or coding being infinitely more accessible and convenient because of AI (No one is willing to help me well with my messy Python code on stack overflow like a tiny LLM is lol), OCR, translation in general, protein folding too? And this is just off the top of my head. Still, this is an ever projecting field, and even if we have positive results right now, we should definitely be careful of conclusions such as "overwhelmingly effective artificial systems that are not aligned with human values," that is if we care about still existing in the way we do then.

The holodeck helped humanity? Have you seen Star Trek?

One step closer to AI+FDVR virtual girls.

better pfps/avatars for them in general

You'll be able to have many waifus at once.

Go find a new technique and come back to us

Why would you expect that? Btw, I've been able to generate good images from SD models relatively straightforwardly. Perhaps you need to develop your skills with the available tools?

How do you figure that a 8B parameter model would be easier to train than a 3.2B parameter model?

@@miriamkapeller6754SD 3 is just built different

It‘ll be free when it‘s fully released

Yea, right. It's just a spam.

...? The model is in preview. It will be released for all, probably in a few weeks.

@@miriamkapeller6754how do people not get what a waitlist is?

@@miriamkapeller6754 yeah in those weeks a better model would already appear... why bother learning this if you can't use it... why gatekeep?

waitlist only means there is a list to wait on. It has nothing to do with its status as a free release