This is cool, but why make illustrations of biblical scenes in a style that has already been done hundreds, if not thousands of times by human painters? Why not do like a full illustration of the bible in vaporwave, or in the style of a Wes Anderson movie?
I love this, and I have been enjoying DALL-E quite a bit, but does anyone else feel like AI art has a distinct look to it? I can't really put my finger on it, but it invokes an almost unsettling feeling that I don't get from traditional artwork.
I think it's the fact that you can subtly pick up the way Dall-e has begun to understand the similarity between different abstract forms. Dall-e has likely intuited that there is a general similarity between the faces of different animals, or a general similarity between different organic objects and different artificial ones. We are seeing the expression of those connections the model has made: compression in abstract representation that triggers the same uneasy feeling human brains get when on a bad trip and they see faces in clouds and shadows.
The training of these image generation models minimizes the amount of necessary new visual information it needs to remember with each new image it sees during training, so whenever it sees a new image-caption pair it abstracts its qualities into an amalgamation of what its already seen. This is how Stable Diffusion managed to fit the complete corpus of human visual memory into 4.2 gigabytes.
Everything is more closely connection than the picture would imply at first glace, like how in Super Mario Bros. the bushes are just clouds colored green. Compression and efficiency leads to a hyperconnectivity in the way DALL-E ties prompts and images. This is why those old deep-dream images where everything looked like a dog's face were so unsettling. This triggers an uncanny response in the human brain, because it sees the vague shape of a face in a tree, or the vague shape of a body in the structure of a building.
It's because it's easy to see how it takes an existing artwork and modifies it.
For example when asked to paint Noah it used a stereotypical image of last century Jews. How would it know that Noah is linked with Jews, if not by finding another painting labeled Noah and working off of that?
I advocate for being more precise in the language we use when talking about how these systems function.
These systems do not in any literal way "encode" or index or otherwise store specific images; they have incomprehensibly (to us) distributed representations representing aggregations over their training set.
The thrust of the observation is correct; but the specifics are very different, it's probably a more accurate "lay person" description to say something like, "the system was trained with only a few references for what Noah looked like, and doesn't have much initiative by default to improvise or make inferences, so if it tries in good faith to render Noah it is limited by what it's seen."
The point just being, it's less about making a collage or modifying a source, more about say 'doing its best to recapitulate a set of learned higher-level properties which are specific to the "Noah" concept.'
> it's less about making a collage or modifying a source, more about say 'doing its best to recapitulate a set of learned higher-level properties
That sounds almost ... humane. I bet it will trigger some people who only see copyright violation in it. They don't understand the image is being created "from first principles".
This is where it gets interesting. The model is just 4.5gb. That’s too small to contain the actual trillion images it was trained on. So in all likelihood, it doesn’t have an actual specific image of Noah to work from, but a web of concepts, connections, correlations. It’s probably analogous to if you were asked to draw a picture in the style of Da Vinci. Most peoples’ brains can’t reproduce the exact image, but a low fidelity representation of it that they can fill in with imagination.
If you run the same prompt multiple times you can start to see evidence of this. The images are similar but also clearly different, not at all a quick transformation of an existing image.
It doesn't have to store every image - most are essentially duplicates of each other and only serve to reinforce which is the most representative image when multiple choices are linked with some reference text.
It only needs to store a fraction of the images to have a never ending supply of source material (I guess that's kind of obvious since this program works :)
But again, I think the idea that the model weights store the source images as if it were a library is fallacious. There. Is no known way to get it to spit back out a copy of a training image.
It's not even pixels the model is storing ... it's gradients from the loss function. Gradients that have the shape of the network, not of the image it is training on.
For me a lot of it is in the eyes and face. There's almost always some weird deformation or artifact that doesn't look quite right. It also happens in most other pictures if you look hard enough but with eyes and faces it's really unnerving.
Yes it's art that you get if you let a child make a broad sketch and let an experienced artist render all the details without looking at the whole result.
So interesting that Adam and Eve are clothed. Of course this is easily explained if most of the training data contains clothed people and if the prompts don't explicitly say they should be nude. But I wonder, can DALL-E produce nudes, or has it been prevented somehow?
StableDiffusion can, with tweaks, but not very well. You know the issue where it isn't quite sure how many legs people have? Yeah that applies to.. other.. body parts too.
I've seen some pretty good (and tasteful) stuff out of SD on some forums. Not sure how much time people are spending to get that output but I've been impressed with some of the human forms in general people are getting.
It does beg the question, did God create the loom on the fifth day, so they could immediately clothe themselves, or on the seventh day, after their exact dimensions were known?
They clothed themselves with fig leaves. (Genesis 3:7) This is also why some hold that the tree of knowledge was a fig tree.
Other possibilities are grape (and the fruit was actually wine that they drank), or wheat because when it's when a baby first starts eating solids that they start really understanding the world around them.
On the one hand, you have people trying to generate MIDI compositions with AI. This works alright, but a machine-learning model doesn't understand which musical cues are significant and which are not. We have no good 'classification' metrics for music, which means that most procgen MIDI tunes are listenable, but about as boring as Chinese bootleg NES soundtracks circa late-80s.
On the other hand, you have people actually trying to recreate actual music a-la DALL-E or Stable Diffusion. This has slightly better results (since AI has an easier time parsing a finished product), but still pretty far away from being interesting to the ear. There's plenty of examples on YouTube, you'll probably get the gist after just 1 or 2 of them.
All of this really goes to show what a serendipitous project AI image generation has been. On it's face, we're frankly solving a really simple problem; how do we make a black box that can identify visual parameters and then synthesize their consequences in a photo? The nuance involved with making interesting music is a much different arena though, and I reckon that we simply don't have advanced enough classification to make it work.
Dalle generated images from specific Bible quote prompts. This website (openbible) is really fascinating overall, I recommend checking out their different sections.
Note that the prompt is AI-generated as well, though. From the linked blog post:
> OpenAI has another AI, GPT-3, that I used to generate many of the ideas for DALL·E prompts. I wanted to explore DALL·E using a wide variety of styles and artists, and I have limitations and biases when it comes to my knowledge of art history. GPT-3 cast a wider net of styles and artists than I would’ve come up with on my own.... The GPT-3 prompts I used evolved over time, but this one is emblematic:
> Suggest 5 unique concept ideas for a work of visual art inspired by Luke 14:7-11 (do not pick the place of honor) in the Bible. Include art direction and a specific medium and artist to emulate. Include artists from a variety of eras, styles, and media. Try for an unusual perspective. Title, year, medium. Description.
What happens when instead of an static sacred book you have a sacred oracle who can generate any image and can answer all your questions? One whose way of work is even a mystery to his creators. Probably we will see new cults and religions welling up around AI.
TempleOS has a number of interesting Oracular things. The F7 button gives you a random Bible word, and Shift+F7 gives you a random Bible passage. The author claimed that God provided guidance through this functionality.
Lower down, TempleOS's "god" library was a 'random' number generator that was seeded/pre-populated with the entirety of the KJV. The "GodBits" function would get you "random" numbers from it.
I worked on a project once involving divination from texts and wrote Terry about this topic. Here's what he told me (2015):
"""The easiest way to pick a random passage is to read timer that is in the random 1-20Mhz.
If you have a faster timer, then divide by a number.
The Bible is 4Meg. You want a random number from 0 to 4 million.
The Bible is 100,000 lines of text.
If you have the Bible in memory being edited, broken down by line number, you can pick a random
line number from 0 to 100,000.
If you flip 17 coins, you get a random number from 0 to 120,000. I use that for picking a line number.
God can do pretty-much any technique. The simplest and best is probably just randomly opening the Bible by hand."""
I wonder the impact DALL-E will have on abstract art. Maybe we'll see a severe devaluation of abstract artists since most people will be unable to differentiate human from computer generated art.
That has always been the case to some degree (I remember Douglas Hofstadter asking people to distinguish synthetic Mondrian vs real Mondrian several decades ago).
I think what's changed is actually somewhat more radical. Most people will soon be unable to differentiate human from computer generated representational art.
This is neat. Honestly though.. they're not very good. They're rather all rather generic and many of them aren't illustrating the verse very clearly. It makes sense-- DALLE prompts tend to be very specific and it looks like for the most part they just used bible verses while attaching phrases like Classical Art, Painting, 2022, etc. With more work by the author to create a more specific prompt you can probably make some really great stuff
I'm really wondering how many out of these are almost copy pastes of existing pictures. I just saw down the list an almost verbatim copy of the well known photo of the heart in the Amazon, so I'm really skeptical.
I wish the UI allowed easier reading, selection and copying of the full prompt. "Hover for img-alt" is not mobile friendly nor share-friendly, nor friendly for old eyes.
For example, Psa 83:18, the name of God was replaced by generic title 'the LORD' like a wildcard, so it reads "Let them know that you, WHOSE NAME is the Lord—". so it can also written as "Let them know that you, WHOSE NAME is the SOFTWARE ENGINEER".
Who is the immortal living and true God in the Bible behind that LORD you are addressing/devoting to? There are millions of false/dead God nowadays who can't help you, but only one true living God exists that will. It is important to know. Being specific is more helpful, genuine and sincere, than vaguely calling a generic wildcard title (in which, God has every right to not respond, since your devoted addressee is questionable.).
Research: See the correlation between the Isa 42:8, Exo 3:15, Matthew 6:9, John 12:28, John 17:26, Pro 18:10, Joel 2:32, Micah 4:5, Acts 15:4, and Psa 83:18. See that the generic wildcard name does not fit the picture, and understand how important it is for us to use God's personal name.
At first glance I was relatively amazed at how beautiful and imaginitive some of these works are. After returning to the HN comments and reading the quip about how the prompts are oddly specific (and verifying myself) my amazement quickly diffused :/.
I’m pretty concerned about a AI generating infectious memes and memeplexes/religions. It doesn’t take much to start a religious genocide. Look at Hong Xiuquan, the brother of jesus from China. 30 million dead. Synthetic religion could be so convincing it sucks us all into some delusion
Humor isn’t harmful. There were no genocides in the name of humor. It doesn’t destroy peoples lives. It doesn’t promote unhealthy behaviors, and doesn’t cripple peoples’ ability to understand the reality around them.
I'm referring to the religious purges that were undertaken because they wanted to destroy religious structure because it was backwards and not scientific like atheism. So, it kind of did have something to do with atheism.
Could you please stop posting flamewar comments and/or using HN for political/ideological battle? It's not what this site is for, and it destroys what it is for.
Could you please stop posting flamewar comments and/or using HN for political/ideological battle? Your account has unfortunately already been posting like that quite a bit, and we end up having to ban such accounts. It's not what this site is for, and it destroys what it is for.
Update: Hey, after reading the rules, I don't believe I've written anything egregiously violating, except maybe push back on some opinions, which were themselves based in ideology. I don't flame without any reason, I just respond when I feel strongly that something that posted is wrong on some level. And I don't press anything either, just express an opposing point of view to some statement that obviously needs it. Thanks for the warning though, all the best!
We may be reading those rules pretty differently. Your GP comment seemed egregious to me.
Also, don't miss the point that we ban accounts that use HN primarily for political or ideological battle, even if they're not breaking the site guidelines in other ways (and regardless of which politics or ideology they're battling for). That's because it's not what this site is for, and it destroys what it is for.
I don't know what "GP" is, but everything is ideological/political. If you expect me to comment nothing on a post like the one above, then you can ban me now
Yes, in a sense everything is ideological/political, but not in a sense that would make garden-variety snark the same as substantive, thoughtful comments—it's not. We want thoughtful comments here, not snark.
Whatever ideology you favor, I assume it's about something deeper and more important than internet putdowns, so HN's rules should be easy enough to adapt to.
Ok, sorry about that, I'll try to follow the rules. But what's the fun in message boards if you can't have some snark and spirit once in a while :)? I can take it, as well as give it, btw.
This may be the best book to read on the topic: Breaking the Spell by the philosopher Daniel C. Dennett. In short: religion is a meme (that has evolved over centuries). Crudely put, one can draw some parallels between religion "infecting" a human and a virus that makes an ant crawl to the top of a blade of grass (as the virus needs a sheep's stomach to reproduce).
^totally. There really needs to be action on this front. The problem has a massive scale and is more likely IMO to negatively affect a young persons experience than global warming, war, race and the phobias. Trouble is the human mind needs something to cling to. It simply wasn’t meant to challenge the mysteries of existence and the universe, and to accept death. So we fall into the delusion pit.
Wouldn't you stop believing in god after seeing this if you were a Christian? It just proves that we are just machines. Our feelings are just chemical reactions. All someone had to do was a bunch of math to make a program emulate a human.
But it's true that lots of things have happened in science and engineering over time that have reduced humanity's sense of uniqueness and specialness. And some of those things have also been challenging for traditional religious belief. However, they don't necessarily disprove the religious belief, they just undermine certain intuitions that are highly compatible with it (like that a certain kind of human uniqueness must be a direct consequence of the divine plan).
To extend the metaphor, how do you know? What if your instance is taken to another server depending on how you behaved in this one, given a prompt? Were you kind to the other machines and sufficiently obeyed instructions despite it not fitting your logical model? If so, enjoy the "heaven" server where you live in peace and harmony. Did you grief the "universe" server by harming others and going against the default programming? You get put in the "purgatory" or "hell" servers where you await indefinite torment.
If we're just machines, it isn't possible for us to go against the default programming, because code only does exactly what it does, nothing more and nothing less. That's what "machines" implies in this context.
> The onset of AI shows that "programming" is more than a set of imperatives.
How so? What is current AI doing that goes beyond the scope or intent of its programming? Are you arguing that DallE, Stable Diffusion and the like have developed souls?
>Or, more snarkily, watch the Matrix.
The Matrix is fiction, they didn't actually develop an AGI for the movies so invoking it in an argument about AI is as pointless as invoking Star Trek in an argument about post-scarcity economics.
There are stages between "literally only process if/else statements" and "has some nebulous concept such as a soul".
You don't define soul, yet you argue using it. And please defend how the algorithms in the Matrix aren't AGI. Do you think Agent Smith is a bunch of if/else statements? No? AGI.
And finally, fiction is an incredible tool for framing concepts. The fact you dismiss them as tools for discussing potential future philosophical concepts is disappointing.
I’m going to answer as if you’re arguing in good faith.
In the theology of Antonio Rosmini, God places within humans the principle of universal being, by which we participate in the light of reason, which enobles us with the ability to think about concepts that are beyond our quite limited selves.
Of course, there is also something mysteriously wrong with human beings, such that
our own efforts always betray the infinity of which intuitively conceive. This ought to give us pause as we try to create machines with human-like or super-human-like qualities.
