The system uses just a few watts of power and keeps neurons alive for up to six months, according to the company’s website.
That's a major point there, cool!
The biggest bottleneck with computers (especially quantum) is the energy consumption, which can be elevated by mixing it with chemical/biological materials.
@SumZero yes, a "chemical" solution might be more accurate than "biological". Although I think there are advantages with using more advanced 'systems' like brain-type (or stem) cells, than just a mix of chemicals and trying to invent the wheel again. Especially because it's (too) complicated to properly understand, so better just use it (like quantum physics in general )
My take on it is there HAS to be a better way than using our poorly made 'best way evolution stumbled on, that we know of' way of processing sensory data and compiling it into thoughts.
I highly doubt of the endless ways to do so, evolution by process of trial and error picked the best one. It just picked the most widely applicable one across a wide range of animals that share our tree lineage branch of clustered cells gathered in one central location connected to a spinal stem to process all body and sensory data. (even octopuses [octopi?] have us beat on that one).
If we try to force it to replicate our/animals way of processing data, we are using a pretty poor and cumbersome way of trying to make something that thinks. It's just the best we know of. Akin to a caveman realizing a broken rock works better to clean hide than a smooth one. Not the best, just the best HE knows of so trying to replicate it in automated machinery to do it using sticks tied to cracked stones being pushed by dismembered hands (replicate how I do it, with things like I have to do it).
Stormwolf - "Who cares about some racial stuff, certainly not the victims."
- Democracy Dies in Dumbness.
- Watching people my age grow from cynical youth who distrusts and dismisses the older generation, into cynical old people who distrusts and dismisses younger generations.
I was speaking more on how we should approach it, not on what we understand to do so far.
That if we are going for AGI, we shouldn't aim for it to be based around our biological makeup. As most our genes in our cells are not the ones designed to do what we are aiming for it to do. Most are for traits to survive, or incremental survival advantage, so why make that the foundation to start from?
we share 99% of our genes with bonobo monkeys. So we would be leaning on that 1% of all our genes that make us 'intelligent' to make it intelligent. Seems highly wasteful. Start ground up learning\making\developing a purely conceptual matrix that replaces that 99% of "Be born, live to make more of us, protect them long enough, then die" coding with something actual useful to it.
And skip the biological part all together. That part is error prone as hell, delicate, environmental condition vulnerable, and overly complicated for just the concept of intelligence. You got to keep a lot of things 'alive' (chemical reactions stable) just so it can exist to DO its thing.
Much like Molecular Biologists and Biotechnologists, they only manipulate and tweak bacteria genes to 'do' a thing, because we don't know how to make things that do that without needing a whole bacterial 'fluff' to store it in.
If they could create it without the bacteria it would save a LOT of hassle. (Not that we should need the bacteria to do it, we only know how to trick simple bateria into doing the end result we want)
Stormwolf - "Who cares about some racial stuff, certainly not the victims."
- Democracy Dies in Dumbness.
- Watching people my age grow from cynical youth who distrusts and dismisses the older generation, into cynical old people who distrusts and dismisses younger generations.
Last edited by SumZero on Fri, 5th Sep 2025 00:56; edited 1 time in total
Thinking about it, my point is much like why we are 'stuck' in binary for computers:
Binary wasn’t chosen because it was elegant it was chosen because it was practical for the hardware of the 1930s–1950.
And it progressed to the standard, so 'stuck'. memristor arrays, ternary logic (three states), or quaternary could possibly be MUCH better if we started with that back then and worked through the tech to modern day.
And its analogous to me if we start with human brain cells for AI (or biological), we are setting a limiting path to be 'stuck' in (jumbled with a lot of evolutionary genes not useful for our intent), as we advance it, than unexplored possible better alternatives would offer.
And if we go far enough, starting over would be going backwards, even if the alternative could do better in the same amount of time (Reversing binary for computers would ruin what we have and be worse starting off, and could take a few decades to catch up and pass where we are now, even after 100 years of binary)
Stormwolf - "Who cares about some racial stuff, certainly not the victims."
- Democracy Dies in Dumbness.
- Watching people my age grow from cynical youth who distrusts and dismisses the older generation, into cynical old people who distrusts and dismisses younger generations.
I was first thinking its click bait but this is very intriguing (about the death of an Open AI employee who called them out, then found dead and with signs of struggle, a wig was found, blood in two different rooms, overall weird).
Altman's answers are not... great.
Edit: oh, the medical report. An absolute joke. It says he took alcohol and GHB and in insane amounts.
As a former user of GHB and GBL, its lethal with alcohol + its super-super fast acting too, there were so many times i took a little too much (you take it in ml, very small dosages) and i was out after less than a minute (you just fall asleep and you go directly into REM-sleep, combine it with alcohol you go into a coma, then death because you will stop breathing).
Why take a lethal combination of drugs and also commit suicide with a gun, likely he wouldnt have the strength to push the trigger at that dosage, or would have been conscious enough to do it
Despite the malfunction, even if it worked this looks gigaretarded.
Just open your phone, look up the recipe and start cooking, maybe? Useless technodick bullshit
boundle (thoughts on cracking AITD) wrote:
i guess thouth if without a legit key the installation was rolling back we are all fucking then
blaming it on the wifi? They really believe the audience is that retarded to think the AI's algorithm has anything to do with the wifi connection.
Silicon valley is out of ideas, hardware plateau'd, everything coming out these days is another attempt at reinventing the wheel but most importantly creating another bubble.
The problem is: The more talk there is about the Gartner Hype Cycle, the more people are aware of it and adjust their behaviours accordingly, the less likely it is to happen. It's like a reverse self-fulfilling prophecy.
Since everybody expects we are now in the beginning phases of the "trough of disillusionment" they invest accordingly and the curve progression will break down again.
It will either blow down past the bottom and fully crash or bump back to the hype cycle very soon. It's not going to go according to the curve when so many people know about it.
This is the 90's internet boom again.
Things are getting silly. Like Nvidia investing in openAI so openAI can lease Nvidia GPU' with the money Nvidia invested. While openAI hardly makes any money. So either they know something we don't, or it's all going to waste and there will be a an AI crash like after the first internet boom.
It's more a matter of finetuning AI and learning how to use it, but AI is still the next big thing. Either in 5 years, or 15, or more, who knows. The potential is there.
The AI Backlash Is Here: Why Public Patience With Tech Giants Is Running Out
From ads scrawled with graffiti to online comment sections filled with mockery, the public’s patience with AI-generated media is starting to wear thin. Whether it's YouTube comments deriding synthetic ad campaigns or scribbled in Sharpie across New York City subway posters for AI startups, the public's discontent with the AI boom is growing louder.
As AI spreads, public skepticism is turning into into open hostility toward its products and ads. Campaigns made with generative AI are mocked online and vandalized in public. Friend, a startup that spent $1 million on a sprawling campaign in the New York City subway with more than 11,000 advertisements on subway cars, 1,000 platform posters, and 130 urban panels, has been hit especially hard. Most of its ads were defaced with graffiti calling the product “surveillance capitalism” and urging people to “get real friends.”
"AI doesn't care if you live or die," reads one tag on a Friend ad in Brooklyn.
The truth is: Slop already started before AI, AI just accelerated the process. It really took off with those creepy Youtube Kids CGI videos that were handmade back then.
IMO I don't see his point? Or nor making a new point. Just the tool changed, the problem with the result already exists for over a decade.
It's like that now to me without AI. 99.99% of stuff on youtube is low effort slop, and 100% of Instagram and tik tok is slop. What/who made it won't be why I don't wish to watch it.
Same argument could be made for them before AI:
What happened to the funnel of video content took expensive equipment, talent, effort and a crew? Now anyone with a phone/camera can film themselves doing shit and post it (And they did, and do).
Im sure there is 10s of 1000s of youtubers with 1000 hours each of stuff, I would consider no better than AI slop, but just made by 'hand' now.
The Mr. Beast example he gave highlights it to me.
Before AI: clickbait slop of him doing stuff. After AI: Clickbait slop of him doing stuff? What's degraded by adding AI? His content wouldn't get worse, just faster. And if I don't watch his slop, he can make 1 a week or 50. Still slop I won't watch.
The way I read his concern anyway: Human made slop, somehow has more integrity than non human made slop. It's the effort put into making the mostly unwatchable slop that counts.
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
I still can't believe how real looking shit that Sora is outputting in video form, its truly something.. I mean it doesn't look CGI at all, it has also evolved so damn fast.
Video and image generation also proves ai is real (non human) intelligence and not some sort of statistical algoritm that is just replicating stuff using math. The models had to learn physics and a lot of cultural stuff just by processing images and video’s. Learning how our 3D world works and learning this from 2D objects. In art, it took humans centuries to understand perspective. And that’s the oppositie: learning how to make a 2D picture from a 3D exemple.
It proofs that ai is not just a trick, it will outperform human intelligence in fields given enough time and development.
Video and image generation also proves ai is real (non human) intelligence and not some sort of statistical algoritm that is just replicating stuff using math.
No it doesn't.
You just have no idea how it works so = magic.
It's like saying someone figure out how to change a simple switch box to something more than a machine, to have emotions and feelings because they made this:
Spoiler:
.
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
I don’t understand your point.
Ai models have knowledge about the real world and they learned that from 2D images and video. Otherwise they can’t make that content. That is intelligence at a superhuman level. It does not proof ai has feelings or is conscious, that’s something else.
Again no they do not. You are using 'knowledge' in the wrong way.
It would be like saying google 'knowns' what spiders are because you can ask it about african spiders and it can direct you to pages of only what you asked for. (Sort of).
They have zero intelligence, or understanding of perspective, or the real world. They replicate pixel colors from other indexed images, based off a network of linked matrix of similar images with similar tags. it's literally mathematical probability of pixel arrangements based off a base seed to produce a grid of dots (pixels) that closely resemble (to you) the subject matter you prompted it with. To it, the math works out to a higher score of being 'like' the other pixel color arrangements in images tagged with the words you gave it.
Its what the 'diffusion' in stable diffusion means:
Start with random noise, and shuffle pixels towards a 'mashup' of all the pixel arrangements that was indexed using the given tokens using a "higher score is better, lower score worse' mathematical system. It what the ksmapler does. A Ksampler is a mathematical convergence algorithm.
Why you have so many samplers, and schedulers to pick from. We have found dozens of different ways to match the 'black box' convergence (unless you are using a non-convergent sampler. but that's a whole different math discussion).
It has no idea 'what' its making, or ideas at all. Why GPUs are good at it vs CPUs. They are amazing at parallel maths, CPU's are not.
And why I can give you a prompt, a size, and seed. And you can make the exact same image as me, on a different machine, at a different time, with systems that never have talked to each other or ever been connected in anyway: It's all math.
Not trying to be mean or snarky. But it works in not even remotely the way you describe it.
I think you might be confusing when hearing: We don't know how it works with a wrong meaning. It means we can't predict what it will produce each seed or know the matrix path it will use for a given input. We have a general idea of why it works. HOW and WHY are two different words.
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
In fact the AI is a black box. But you can't make a video without knowing something about physics, lightning, etc. It has something the model has learned from observation. This can't be generated by "chance"
There is a research paper (can't find it now) for instance that looked at how an AI draws a car. And the first step that AI took was making a 3d outliner showing it "knows" were that car should be in a 3d space in a structured proces.
The basis of an AI is of course choosing the most plausible token. But AI is much more complex then that and it's just the basis, the complete neural network is more like a real brain. A complex and messy piece of knowledge and skills that has somehow learned do the things it has been asked to do. See this article for instance about neural networks in a language model.
As that is not in anyway, how it makes a car in an image. There is nothing, I am being absolute here in certiantly and terms. That no Ai image model uses, knows, understand, or even has a conceptual matrix of 3d positional locations. Be it SD, SDXL, Illustrious, FLUX, NooB, or WAN or such based. (the foundational base types of all image models)
I have used 100's of local and server cluster AI image models, and made probably terabytes of images. Not one image or video model has any concepts of 'placing things in 3d space'.
It wouldn't be image diffusion, if it started with a 3d volumetric sketch of an object it 'knows' as a car, and how it fits into a world.
The only ones that do are not image models, they are 3d modeling/svg production Ai models. And they do not understand 'what' they are doing with any intelligence they are parroting/mimicking variations of what they learned is most probable mathematically.
I cannot convince you it's not intelligent or 'smart'. Only you learning of it inner workings on your own can help you realize it.
Edit:
The link you gave might need to be read slower, with lookup of references it offers?
It starts out with stating it's using quotes to point out "We are changing it to words that make sentence flow better. that neurons, is a replacement word for floating points numbers in math". To help 'common down' the explanation. in NO way are they saying it IS a neural network. They are comparing it to one for sake of a concept close enough you already understand.
Much like telling someone that never seen a plane: they have wings that help it fly, much like birds use wings to fly. In no way does that mean planes are a species of bird. Or that they are conscious like a bird. It's simply going: here is a thing you know, it's close enough in analogous intent to get a concept across. but it's NOT that.
Or like if you was teaching someone math and used apples as an example of addition.Then they erroneously assuming math has something to do with apples.
All it is saying is 'Because the knobs are not labeled that we tweak, we do not know what to expect in the math it uses for output". That it has 'math' knobs. We just don't know what they change..but we know it's to do with math of its process.
I assume (wrongly maybe?) you see words like 'mind' and 'learning' and apply anthropomorphism to what they mean out of context of what they mean them as?
Like 'thinking' SHOULD be 'calculating'. But thinking is more tangible to grasp to the end user the process used of forming a sentence. 'Calculating' doesn't seem like a correct word to people using it, for making sentences, so they use 'thinking' as it makes more sense to average joe. (it also helps sell it to average joe. less 'computery' and more 'it is thinking, like you do, to say something')
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
Last edited by DXWarlock on Tue, 7th Oct 2025 19:45; edited 1 time in total
I realy can't find that article about 3d spacing, but it's how it works. At least in that study, the model started with outlining an object in 3d space, starting with outlining what is in the front, middle back and working the details out from that point. And then adding more details. It's a structured proces that requires knowledge of from our physical world. Otherwise the lighting for instance would not make a lot of sense for us. The model has to know that there is a source, that it effects colours, that it can reflect, etc. It does not understand what it is doing, but it does know how to do it, at least in a good enough way to fool us.
For me the proces these models have gone through is a sign of intelligence: being able to learn how use abstract concepts to complete a task.
It is not how it works. I will bet both my cars and my house on it.
The way you explain it is kinda right. But the concept of what it is doing is wrong. They may have been using analogies to convey concepts, and you took it as descriptional?
There is 8-12 (or 24 depending on the model base type) blocks of nodes for step instructions. Some for composition, some for color, some for definition, and some for final resolve of the 3 above block layers. But it always, always, works in 2d pixel grid space. Its flipping colors, brightness, contrast, etc on a grid of pixels to mimic the 'same-same' it has stored as millions of images sharing tokens you feed it.
I wrote a whole civit article on it to help people make custom models for tabletop fantasy images with a realism look (as realism models suck at orcs, goblins, dwarves, elves, etc) for model merging and what blocks do what, so you can pick which parts of which model control which part of the image process:
Spoiler:
It processes blocks in a linear(ish) pattern of noise to the final image, each impacting the process along the way, and which model it leans towards for that step in the process of noise -> final image. For tuning which model in the merge does which steps these are the blocks to give attention to:
In this example Model A is a anime/cartoon base for exaggerated poses and fantasy outfits/creatures. Model B is a realism one for final upscale textures and realism.
1. Blocks: Base, IN00–IN04
Function:
Define the scene layout (composition).
Handle poses and anatomy for characters or objects in the scene.
Impact of Weights:
High Model A weight: Keeps the structure anime-like (e.g., exaggerated poses, stylized proportions).
High Model B weight: Makes the structure more realistic (e.g., accurate human anatomy, balanced proportions).
2. Downsampling Path (IN05–IN11)
Step: These blocks extract increasingly detailed features by reducing the image resolution (downsampling).
Blocks: IN05–IN11
Function:
Refine mid-level details, like shapes, textures, and shading.
Begin introducing style features (e.g., anime or realism textures, basic shading patterns).
Impact of Weights:
High Model A weight: Keeps mid-level details stylized (anime-style texturing and shading).
High Model B weight: Introduces realistic textures and consistent shading.
3. Bottleneck (m00)
Step: This block, also called the middle block, connects the downsampling and upsampling paths.
Blocks: m00
Function:
Encodes the global style and coherence of the image.
Balances the influence of broad structure and fine details.
Impact of Weights:
High Model A weight: Preserves anime-like global style.
High Model B weight: Shifts the overall style toward realism.
4. Upsampling Path (OUT00–OUT03)
Step: These blocks begin refining the image by adding detail as the resolution increases (upsampling).
Blocks: OUT00–OUT03
Function:
Handle mid-level textures and shading transitions.
Ensure style consistency between the structure and details.
Impact of Weights:
High Model A weight: Retains stylized textures and shading patterns.
High Model B weight: Adds realistic texturing and tonal depth.
5. Fine Detail Refinement (OUT04–OUT07)
Step: These blocks refine the high-resolution details, like facial features, skin textures, and intricate elements.
Blocks: OUT04–OUT07
Function:
Enhance faces, fine textures, and other small details.
Add realism or stylization to critical focal points (e.g., eyes, skin, hair).
Impact of Weights:
High Model A weight: Faces and details remain anime-like (simplified or stylized).
High Model B weight: Faces and details become highly realistic.
6. Final Touches (OUT08–OUT11)
Step: These blocks handle the final stages of image generation, adding shadows, highlights, and other finishing touches.
Blocks: OUT08–OUT11
Function:
Enhance lighting effects, shadows, and overall rendering quality.
Ensure the final image looks cohesive and polished.
Impact of Weights:
High Model A weight: Keeps stylistic shadows and artistic flair.
High Model B weight: Adds realistic shadows, highlights, and tonal balance.
Summary of Block Roles in Image Generation
Base, IN00–IN04
Define composition, pose, and basic structure.
Feature Extraction
IN05–IN11
Add mid-level textures, shapes, and shading.
Global Style
m00
Blend and set the overall style of the image.
Detailing
OUT00–OUT03
Handle mid-level textures and shading transitions.
Fine Refinement
OUT04–OUT07
Add realistic or stylized high-resolution details.
Final Rendering
OUT08–OUT11
Apply shadows, highlights, and polish the image.
UNET:
- model.diffusion_model.input_blocks.0 to 8
- model.diffusion_model.label_emb.0
- model.diffusion_model.middle_block.0 to 2
- model.diffusion_model.out.0 and 2
- model.diffusion_model.output_blocks.0 to 8
- model.diffusion_model.time_embed.0 and 2
CLIP:
- conditioner.embedders.0 and 1
VAE:
- first_stage_model
No where in any of it, in any image model is it outlining a 3d model, and positioning it in 3D space. Nor know what the 'front/back/side' of the car is. For the other parts about knowing reflections, and volumetric concepts of accurate shading based on real world physics understandings (or what have you)..that is not a 'thing' and not sure where you got that.
Reflections are made by smashing together millions of pixel info from images with reflections to paint pixels that have that 'pattern'. It has no idea what 'shiny' is as a understood concept. And because of that, it gets them wrong...a lot...a whole lot.
Go run say Auto1111 with preview steps set to 1. You can see realtime how it does each step.
You might have saw one on 3d compiling/Modeling of objects. But that is not even remotely the same model modal as image generation.
friketje wrote:
For me the process these models have gone through is a sign of intelligence: being able to learn how use abstract concepts to complete a task.
It needs more than that. otherwise advanced and abstract use cases of Excel and Photoshop have intelligence.
You confuse "I don't know how it does it" with "I don't know how it does it, so I will imbue it must do it like me"
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
Signature/Avatar nuking: none (can be changed in your profile)
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
)
!?!?! so me using the indian chineze programmers to do my work wasnt ethically?
