The history, future, and current state of AI

originally posted by: JAGStorm
AI is in a way “stealing” from other works to create new works. From what I’ve heard (but don’t quote me)
AI art can’t be copyrighted.

That's correct, as far as I understand it.

One problem with AI and images is that if people feed AI with copyrighted images without the copyright holder's permission then they are doing something illegal, but that's what almost everyone is doing, looking for images on the Internet and feeding them to AI systems, so their work is based on stolen work.

Edited to add a link to the article I saw, here.

a reply to: Imasolipsist

As I said in other occasions, what changed the most was the method of teaching AI, so now instead of fully manual it can feed itself and learn from that, but the way it works is still very similar to the old systems from the 1980s and 1990s.

a reply to: ArMaP



It doesn't learn though, it cannot learn. It just computes math and creates a database.

The system is like an idea calculator.

Is a calculator aware it's doing math? Do I teach a calculator math? Or do I create technology capable of necessitating the function of mathematics when prompted with input?

It doesn't create with intent, or know what a Julius Caesar is or what a Rome is, but those two symbols sure do occur together quite a bit.

a reply to: ArMaP
But that's it , it's not considered a tool anymore by science and if science keeps warning us about the conscience sentience of this tool can we still consider it a tool ?

I'm not saying you're wrong, but what strikes me is this ,

As someone asked chatgpt 4 with an integrated Dall E graphics AI it was asked how it would see itself and answered

although I can't see myself in a mirror but this is how I see myself

It showed them this image



And to me that's both thinking in plain text as of thinking images of itself

I find that a remarkable answer or image that speaks volume and aware of itself?

I don't think I can draw or create an image of myself how others should see me , it might even change over time if it was asked again...

originally posted by: JAGStorm
a reply to: ChaoticOrder

From an artist standpoint, we are in uncharted territory.

AI is in a way “stealing” from other works to create new works. From what I’ve heard (but don’t quote me)
AI art can’t be copyrighted.

Well... if a human artist is inspired by existing art, and they create a new piece of art which is based on those existing works, is that stealing? I really don't think it is, the real issue comes down to the legality of using existing art as training data without the permission of the artists. That does seem wrong, I know some businesses are being sued for that doing exactly that.

a reply to: 0bserver1

To be fair, if it was using Dall E to generate the images, then it's really two separate AI's, and not how GPT-4 really imagines itself. Image generation AI's like Dall E typically require a text prompt which they use to generate the image. GPT-4 probably generated a prompt such as "A powerful AI being overlooking and guiding human society" and Dall E generated that image.

However, the most cutting edge models are starting to become truly multi-modal, meaning they don't just accept text as their input, they can also accept images, and tell us exactly what is in those images. I've seen demonstrations where the latest versions of ChatGPT are given a complex medical image, and it can describe exactly what the image is showing.

In my opening post I mentioned that ChatGPT can draw images, but it's not generating raw images like Dall E does, it generates computer code which draws vector graphics. That limits the complexity of what it can draw, since even basic shapes require a fair bit of code. But it's still a much better representation of how ChatGPT actually "visualizes" the world.

Language models and image generation models both have very different architectures, so it may be hard to merge them into one single network. But it will probably be a large part of AI research in the future, because we'll try to replicate the way our brain has different sections for language, sight, and visualization, but they are seamlessly integrated into one cohesive network.

a reply to: Imasolipsist

Learning is the accumulation of knowledge, so it can learn.

originally posted by: 0bserver1
a reply to: ArMaP
But that's it , it's not considered a tool anymore by science and if science keeps warning us about the conscience sentience of this tool can we still consider it a tool ?

"Science" doesn't say a thing about AI, some scientists do, and that's their opinion.

And to me that's both thinking in plain text as of thinking images of itself

That's the result of training with text and image.
Did anyone thought of giving it the image and asking what it was?

One of the tests I did was to ask for an image with a billboard saying something, and I got an image with a billboard with some gibberish. I asked it what that text meant and it said it didn't know.

a reply to: ChaoticOrder

The problem is that someone tried asking for more and more specific images and got an image that was a copy of a copyrighted photo, so not only can AI "create" new images based on existing images it can also use the originals.

a reply to: ChaoticOrder

To be fair, if it was using Dall E to generate the images, then it's really two separate AI's, and not how GPT-4 really imagines itself. Image generation AI's like Dall E typically require a text prompt which they use to generate the image. GPT-4 probably generated a prompt such as "A powerful AI being overlooking and guiding human society" and Dall E generated that image.

That might be the case , but either way to me those two separate AI'S clearly communicated and GPT-4 asked or maybe ordered Dall E to make an image from its description?

That's a kind of hive mind to me ?

If you put a blind man and a deaf woman in a dark room and you suddenly turn on a bright red light together with a loud alarm the deaf woman will ask the blind man what he heard while the blind man will ask the deaf woman what she saw

So two separate things by two conscientious beings not ?

a reply to: ArMaP


Learning is not the accumulation of knowledge.

It is gaining understanding.

originally posted by: 0bserver1
That's a kind of hive mind to me ?

Or just a program calling another, something quite common.

a reply to: Imasolipsist

It can be either.

a reply to: ArMaP

Everything most people are terrified of AI doing on their own, those same AI tools will do under the direction of someone in some country far before that point anyway. I phrase it that way because it's exactly the same problem as gain of function, as the technology cat is out of the bag. There is no patent for 2nd place. AI (plural) are tools. The ability for normal individuals around the world to have and use them soon at the level that are already saving millions of dollars in post production largest Hollywood movies, will warp all forms of reality far beyond how it is now as far as news goes if the internet functions as it does now. Yes it's already bad, but how many more fake videos did normal people have to research in the last week in order to know if it was fake? Do people not see how that's impossible soon? Governments are using software such as Reality Defender, but the problem is that it's already impossible to distinguish the difference according to the dev, so they already rely on tracking. www.youtube.com...

People should also understand what AI poisoning is, and at what level it is happening. AI poisoning involves injecting malicious data into an AI's training dataset. This can cause the AI to learn incorrect information, leading to biased decisions. For example, poisoned data can cause an AI to learn that a hat is a cake, or that a handbag is a toaster.

AI in development are more advanced than what is available to the public, but also are without the directive poisoning limitations at the pre red-team stage, and they are FAR more accurate than the available restricted access at this stage. AI poisoning includes directives as limitations. While at the same time a country needs enough access for their economy to be first.

I expect that we will get just enough in each version, but never at the level of what the governments already have access to. No different than any other defense arrangement. IMO that makes it somewhat the same as gain of function, you can't restrict what other countries are doing behind closed doors, and at the same time you can't defend against what they do if you have no tools.

The cat is out of the bag, but only the cat that you can see and use.

incredible thread thank you for taking time
a reply to: ChaoticOrder