They Just Released An Image Generation Tool Thats Too Realistic...

They Just Released An Image Generation Tool Thats Too Realistic...
Prepare for AGI with me - www.skool.com/...
🐤 Follow Me on Twitter / theaigrid
🌐 Checkout My website - theaigrid.com/
Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.
Was there anything i missed?
(For Business Enquiries) [e-mail]
#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

[@DiceDecides]

<a href="#" class="seekto" data-time="736">12:16</a> bro just said "Nature has unnatural shapes"

[@DinoByteSize]

Check your unrealistic subtitles before posting.... Bro

[@fromduskuntodawn]

As a ratio of our short attention spans, and the quality of the photos, they are basically realistic to most people at this point.

[@mynamesjeff69]

Ah yes blocks 1.1 hahaha

[@DinoByteSize]

@@mynamesjeff69 1.1 bro!

[@DiceDecides]

can't wait for the 1.12 release

[@wookiee_1]

<a href="#" class="seekto" data-time="583">09:43</a> The reason adding something like "IMG_0134.CR2" works is because LLMs don't truly understand what they are producing or answering (despite all the "AGI" hype). They are essentially knowledge databases that link variations of prompts to variations of answers. It recognizes the image format in the prompt and returns what it thinks is a similar style (from its training data) under the constraints of the rest of the prompt.

[@marinepower]

every AI model works the same way -- they find commonalities within batches of data. With big enough batches of data it learns that a certain tag (say, IMG_0134) has certain implications (taken on an iphone, probably of nature, etc). As for whether or not it understands, obviously it's not conscious, but at the same time with enough training I don't buy the argument that it doesn't understand. It has essentially distilled down what it knows of reality down into a series of complex algorithms (how weird and lumpy rocks should be, how snow looks, how a river looks flowing around some rocks, what humans are likely to take pictures of, etc, etc). So in some very real sense it has internalized within itself a (fairly accurate) model of reality. So it might not 'understand' what its doing, but it's still accurately capturing and reproducing most of reality.

[@blackstream2572]

@@marinepower The way I see it, most of the confusion comes from the (IMO) fact that AI understanding and human understanding is similar but also very different, which seems to lead to people dismissing anything it does better than humans as algorithms and computers doing what they do best and anything it does worse than humans as proof that it has no real understanding or sentience. So that's my opinion of how we got people deciding AI has no real understanding because it factors the filename into what kind of output it produces, yet meanwhile we have a huge subset of our population that's convinced that your astrological sign plays a major role in your personality and future and that you should knock on wood when talking about potential bad future events 'just in case'.

[@wookiee_1]

@@marinepower All great points, and that's exactly what neural nets are, but I don't believe understanding can really exist without consciousness. Understanding involves some agency and drive, without it, this is just searching for relation and pattern recognition. I've seen this behavior in every other major LLM as well, despite being trained on vast amounts of data already, asking for something that is within the same domain, but novel tends to turn them into chatbots that spew nonsense because they are incapable of understanding. If they truly understood, you could feed them a math textbook and they would be able to distill first principles from it to solve new problems within the same context, but they don't. Instead, they are given billions of question-answer pairs in training, which is what they need to perform a search, pattern recognition, and generation.

[@marinepower]

​@@blackstream2572 That could very well be it. Words like consciousness, understanding, etc, are all very human-centric so it makes sense why there would be confusion. Having said that, I have come up with a few definitions that unify human + ai concepts that I personally find useful. Understanding - the degree of the understanding of a system is the accuracy of its predictions in a given domain. 100% accuracy == perfect understanding. A markov chain might have very low understanding, an LLM a fairly high level of understanding, a human a very high level of understanding, and a calculator + human (together as a system) could have perfect understanding (if we're talking about basic math). Consciousness - any system that is given senses, takes actions based on those senses, and then learns from said actions is conscious to some degree. The fidelity of consciousness is mostly based around the depth of processing and the quality of the senses. Things like self-awareness, thinking, etc, all drop out of feeding your thoughts back into yourself as additional senses. Any being that has any consciousness has qualia (since qualia is the unique way you perceive your own senses -- and any being capable of learning will have their senses colored by their life experiences).

[@marinepower]

@@wookiee_1 ​ I think understanding can exist without consciousness, however your point is well taken that right now all that models are doing is learning a data distribution in a fragile way. Issues with modern LLMs: A model doesn't know what it knows or doesn't know. (no self-awareness) A model doesn't know how to extrapolate (no reasoning, no curve fitting) Still, to me, understanding is mostly about the accuracy of a system within a given domain. It doesn't really make sense to me that a system with 100% accuracy can be said to not understand something. Under this definition, a markov chain might have very low understanding, an LLM a fairly high level of understanding, a human a very high level of understanding, and a calculator + a human (together as a system) could have perfect understanding (if we're talking about basic math). So, in my view, an LLM has decent understanding in most domains, but no perfect understanding in any domain (due to its nature as a neural network trained on statistical patterns). That being said, I could see an LLM agent that writes code for itself to solve problems and then runs said code to have something akin to perfect understanding / extrapolation. We're not there yet but we're getting closer.

[@TrabSr]

Ok this ripped me up at <a href="#" class="seekto" data-time="737">12:17</a> this line. "This looks a little bit more natural, because like I said before nature has these you know unnatural shapes and the randomness is inherent in their original design." Spoken like a true wordsmith there buddy :D. Yes things that have regelaur shapes, those are natural! Not the weird nature all those ugly irregular shaped, BAH!

[@cshepherd3310]

<a href="#" class="seekto" data-time="500">8:20</a> Her earrings are mismatched and neither of them look quite right. That's what gave it away for me.

[@TomaszGumny]

<a href="#" class="seekto" data-time="284">4:44</a> - It IS AI : level of water between 2 central stones is to high; up over the surface behind and left

[@JollyJoe135]

We need more prompt adherence in the space for real. It’s annoying trying to get the model to do what you want especially with longer prompts.

[@LiquidAIWater]

It has to be in short clips as not to suck up too much resources.

[@LiquidAIWater]

The photo of the stream has an error. Notice the water coming off the rocks is all level except one that is elevated, but no terrain behind it to justify this. I'm all about water! 😂

[@armadasinterceptor2955]

At this point all these generators realistic, im not even shocked, and dumfounded anymore😂

[@JonnyCrackers]

<a href="#" class="seekto" data-time="477">7:57</a> Really? You don't see anything in that image that gives it away as AI? Her earrings are different on each ear, the one on the left is looking like it can't decide if it's jewelry or a lock of hair. The one on the right has some kind of weird string dangling from it.

[@snowman3995]

That doesn’t indicate ai, especially since there are actually people who wear earring like that.

[@JonnyCrackers]

@@snowman3995 You're tripping.

[@Albaloola]

Is this just me or has the audio been unbalanced towards the left side for the past few videos?

[@georgemontgomery1892]

@<a href="#" class="seekto" data-time="512">8:32</a> One giveaway is the lady's earrings.

[@meandego]

AI should be able to generate multiple images with the same person in different situations and at different ages. Generating one random image is not enough for me. Cool that it looks real, but so what? I can't scam anyone with one image on OF 😂

[@mike6335]

<a href="#" class="seekto" data-time="519">8:39</a> only the ear ring

[@davekite5690]

"nature has these unnatural shapes..." - are you sure about that ;-)

[@ROBOTRIX_eu]

[@MCWolf14]

Bro uses ai for his subtitles

[@Cory-v4w]

Looks like you can't rest on laurels in an Ai race. It would appear, that my before & after is concisely congruent. Burn that midnight oil & think twice.

[@AI_Superintelligence]

Another one 😂

[@thedannybseries8857]

First!