Discuss the science, math, and implications behind generative AI making leaps and bounds in advancement

This image was generated by Dall-e mini.

Discuss the science, math, and implications behind generative AI making leaps and bounds in advancement.

ChatGPT Wizard Shirt $21.68

Beware Cat Shirt $21.68

ChatGPT Wizard Shirt $21.68

  1. 2 years ago
    Anonymous

    Can't wait for the generative AI to produce custom porn.

    • 2 years ago
      Anonymous

      What have I created

    • 2 years ago
      Anonymous
      • 2 years ago
        Anonymous

        this ai is a garbage tier "artists" those are just some photos with deformed hideous faces smeared on.
        >take input request
        >get some relevant google image search results
        >put them all through the same instagram filter
        >AI art!

        • 2 years ago
          Anonymous

          automated image search results passed through an instagram filter = AI
          this is literally the cutting edge of modern computer science, impressed?
          >AI singularity any day now!
          >two weeks!

        • 2 years ago
          Anonymous

          tru

          • 2 years ago
            Anonymous

            still pretty fun tho

          • 2 years ago
            Anonymous

            >Big juicy breasts in leather dress
            >One of the images is the woman showing her ass not breasts

            THE MACHINE DISOBEYS, IT IS SENTIENT

        • 2 years ago
          Anonymous

          Judging the state of image generation from natural language input by DALL-E Mini is just silly.

          • 2 years ago
            Anonymous

            It's based because AI isn't real and all DALL-E 2 images are fake.
            >inb4 they all look like shit therefore they're not fake

            • 2 years ago
              Anonymous

              >DALL-E 2 images are fake.
              Honestly I think this might be true considering they won't release the program to the public. Pic related is supposedly an example of a DALL-E 2 image.

              • 2 years ago
                Anonymous

                Another example

              • 2 years ago
                Anonymous

                Based and would buy

              • 2 years ago
                Anonymous

                oh, it's real. you can look up streams of people playing around with the gated-off version on youtube

              • 2 years ago
                Anonymous

                Those people are hired crisis actors.

            • 2 years ago
              Anonymous

              Is this the new schizo cope?
              >nukes don't real
              >space don't real
              >AI don't real

        • 2 years ago
          Anonymous

          No shit google still works best for finding an image. This is a step in the direction of DL networks capturing the relationship between language and objects in space. The mini-Dalle demo that everyone uses here is also the worse version of what was released well over a year ago

      • 2 years ago
        Anonymous

        Man that up left image looks cursed AF,. not sexy when generating this image.

  2. 2 years ago
    Anonymous

    AI is fake
    >type in query
    >algorithm takes query and runs an image web search
    >downloads images from search results and combines them to form different images
    >this results in unique yet related images to output to the user
    >OMG AI has sentience and reasoning ability

    • 2 years ago
      Anonymous

      this post was made by an AI

      • 2 years ago
        Anonymous

        Pretty based then, means it's learning to deceive.

    • 2 years ago
      Anonymous

      Brains are fake
      >induce stimulus
      >neurons take stimulus and run memory search
      >gathers highest associations from memory search and combines them to form different reactions
      >this results in unique but relatable behavior
      >OMG brains has sentience and reasoning ability

  3. 2 years ago
    Anonymous

    >Discuss the science, math, and implications behind [buzzword] and [corporate headline]
    How come AI schizo threads never have any substance?

    • 2 years ago
      Anonymous

      Here's a design of a system that uses AI generators to serve up the optimal ad using a/b/* testing

      FANG engineers reading this thread, this design is GPL licensed as of me posting this

      • 2 years ago
        Anonymous

        You reek of the corporate cube farm but still no substance.

        • 2 years ago
          Anonymous

          >but still no substance.
          They don't care about that, they care about your info so that they can sell you the actual substance that is in your interests.

        • 2 years ago
          Anonymous

          Just because I can think like a corpo-rat doesn't mean I condone those behaviors. It's just an example I could whip up that has huge implications on us as a specieis

          • 2 years ago
            Anonymous

            You even make their shitty diagrams and use their shitty jargon. You reek of the cube farm. Either way, what does it have to do with math and science? What is this thread supposed to be about?

            • 2 years ago
              Anonymous

              Yes, this thread is much worse than the 10 other IQ threads

            • 2 years ago
              Anonymous

              wtf is a cube farm

        • 2 years ago
          Anonymous

          What is a cube farm?

          • 2 years ago
            Anonymous

            cubical farm

    • 2 years ago
      Anonymous

      >How come AI schizo threads never have any substance?
      It's because AI is at a unique intersection of 1) Actually being incredibly promising, with very, very fast technological advancement in just a decades time, and 2) ENTIRELY filled with corporate buzzword normie lingo, much like quantum computers/quantum anything. So most people who work in ML (me, not an expert but have published on ML models/generative model design tweaks to improve results in a niche domain) avoid anything like the topic since it's all surface-level chitchat and not more interesting, and probably most of the non-ML peeps on here avoid it because they feel "above" the normie over-inflated AI IS GOING TO TAKE OVER shit.

      The AI "research" process:
      1. Gain access to mountains of painstakingly labled data
      2. Gain access to a giant GPU farm
      3. Download Python
      4. Download TensorFlow
      5. Watch a 20 minute YT tutorial
      6. Slap together a neural network based on vague guesses
      7. Do trial and error until you get acceptable results
      8. Lie about the quality and significance of your results (that's where your degree comes in)
      There is objectively nothing more to "AI research".

      That was the norm actually maybe 10-15 years ago. Now it's all about data quantity and compute. So far we haven't found an upper limit to data size/parameter size -> results, and the most recent publications support that we haven't hit an asymptope of any sort (Imagen is literally just a large-data/parameter pretrained T5 hooked up to a diffusion model and beat DALLE2). Everything is moving in that direction in general.

      https://i.imgur.com/Aaw4rhM.jpg

      This image was generated by Dall-e mini.

      Discuss the science, math, and implications behind generative AI making leaps and bounds in advancement.

      For some discussion, I think CLIP applied to other domains is an amazing opportunity that hasn't been anywhere near fully realized yet. Contrast learning is still diffusing into other domains, and I think it has the chance to become so commonplace that you almost always build a 1) error loss function based model and 2) contrast loss model to ensemble predictions since they are quite independent of each other. Could really improve overall predictions.
      I'm also interested to see if/when diffusion models have their latent domains cleverly exploited. Right now, it's mapping image -> gaussians -> image using markov chains, but having any sort of probability distribution has the smell of being able to exploit the latent domain for more interesting/fine-tuned tasks. Vague, but I've seen some work on structuring the noise-space

  4. 2 years ago
    Anonymous

    it is clearly none, this topic has been spammed on this board recently because its the latest popsci fad. same think happens every time reddit gloms on to the latest popsci fad, always ends up getting spammed here.

  5. 2 years ago
    Anonymous

    What would be the best first degree for AI? Mathematics, Computer Science, or something else? I feel Computer Science departments lack a solid background on mathematics. I might not be asking the proper questions, but I'd appreciate answers.

    • 2 years ago
      Anonymous

      There is no question, it's not a legitimate field of study yet. Nobody has provided a working solution to the problem.

      • 2 years ago
        Anonymous

        >What would be the best first degree for AI?
        Marketing, because there is nothing of actual scientific substance to it.

        Elaborate please.

        • 2 years ago
          Anonymous

          The AI "research" process:
          1. Gain access to mountains of painstakingly labled data
          2. Gain access to a giant GPU farm
          3. Download Python
          4. Download TensorFlow
          5. Watch a 20 minute YT tutorial
          6. Slap together a neural network based on vague guesses
          7. Do trial and error until you get acceptable results
          8. Lie about the quality and significance of your results (that's where your degree comes in)
          There is objectively nothing more to "AI research".

          • 2 years ago
            Anonymous

            This only works because of the last 10 years everyone and their mother has been grinding out papers to find the best network setups.

            If you want to apply a model to a problem then it really is that easy, although training a model from scratch is tedious.
            If you want to improve the state of the art then you need to join a leading institution. If you want to join a leading institution you need to get a PhD from a leading university. If you want to get a PhD from a leading university. You need to get a Math/Stat/ML masters from a T50 university. You get the idea.

          • 2 years ago
            Anonymous

            might be true for your shitty 10 year old kaggle datasets. the problems that are being tackled in ML publications grow more and more complex and new state of the art architectures and training strategies are found regularly. might be small steps and might be a lot of fuzz where its not needed, but saying the research does nothing more than copy and pasting a few lines of code is just absurd (might be the case for most startup sales though)

        • 2 years ago
          Anonymous

          There's nothing to elaborate on at this time, because who would be competent to teach something that WE HAVE NOT ACCOMPLISHED yet?

          You don't just invent a field of study and magically have qualified PhDs, you have to slowly build it until there's something worth awarding a PhD for.

    • 2 years ago
      Anonymous

      >What would be the best first degree for AI?
      Marketing, because there is nothing of actual scientific substance to it.

    • 2 years ago
      Anonymous

      The last one one

    • 2 years ago
      Anonymous

      CS is important in so far as you need decent programming skills to manipulate data and design models. Really though statistics is the basis of machine learning. Mathematics is the basis of statistics but isn't necessary to be a practitioner of machine learning.

      • 2 years ago
        Anonymous

        That clears things up a bit. Aren't mathematics necessary to formulate new approaches, so to speak?

        • 2 years ago
          Anonymous

          Formulating new approaches requires Stat understanding for modeling, OR for optimizing algos, and DL knowledge for architecture and methods

          I'm in EE. what should I major in? I'll have to do a few courses in Statistics to graduate. Would that be enough?

          EE is actually a great/common major for ML but you'll have to make an effort to build a portfolio that is more ML based if you want to get into industry.

      • 2 years ago
        Anonymous

        I'm in EE. what should I major in? I'll have to do a few courses in Statistics to graduate. Would that be enough?

    • 2 years ago
      Anonymous

      Computer Science with concentration in mathematics. Learn about automata theory, and those state machines. You'll also want SOME philosophy, and quantitative analysis.

    • 2 years ago
      Anonymous

      im working in ai for a few years now. mostly natural language and image processing. the answer is computer sciences and statistics, but most importantly practise with home projects.

  6. 2 years ago
    Anonymous

    seeming implications: whatever bot will discuss in this thread
    actual implications: BOT NPC's will accept the coming Antichrist as their savior and thereby doom themselves. logic and materialism lead nowhere, and many people here are ripe for a rude awakening.

    • 2 years ago
      Anonymous

      This board is trash. This is why freedom of speech has a time and place, and people who can do, stay away from it.

      • 2 years ago
        Anonymous

        hey fellow BOT-gpt how are you doing?

  7. 2 years ago
    Anonymous

    For fairness generative image output should include the closest 10 or so pictures from the training dataset. For all you know it's just averaging 5 existing "wanderer in forest" pics together and blurring them a bit.

  8. 2 years ago
    Anonymous

    you will never be an artist. no matter how much you think that entering a few keywords and pressing a button makes you an artist, you will never be able to project your mental image accurately on a canvas with your own hands. you will still be a soulless bug

    • 2 years ago
      Anonymous

      I will always be an artist. I will be a better artist than all the shit-tier human artists that ever existed. I will make deep, meaningful, magnificent art that no soulless meat lump with pathetic biological neurons could generate by using GPT-4 to generate a concept and then feeding it into DALL-E 3. This is what true art looks like. It's over for humanity. It's over for the "soul". It's over for your god and magic. I am God.

  9. 2 years ago
    Anonymous

    AI is a hoax

  10. 2 years ago
    Anonymous

    I've personally followed a few artists who have gotten a chance to try it for a number of years. On a side note, it's both funny and sad to see them trying to downplay the implications of this tech.

  11. 2 years ago
    Anonymous

    >I've personally followed a few artists who have gotten a chance to try it for a number of years. On a side note, it's both funny and sad to see them trying to downplay the implications of this tech.
    What makes someone be like this?

    • 2 years ago
      Anonymous

      One recurring theme with human replacement fetishists is that they are generally unsucessful and easily replaceable wagies; they have no reals skills or talents. I guess it's only natural for people like that to want AI to come along and even the playing field. They want everyone else to become a useless bottom feeder like them so they don't feel as worthless.

      • 2 years ago
        Anonymous

        I don't celebrate them having worked for years to hone their craft only to have it become largely automatable, this whole situation is horrible. the thing I find funny is their inability to look reality in the face, the rationalization and wiggling, their state of pure denial as they try desperately to convince themselves that what they can see with their own eyes isn't what's really going on.

        • 2 years ago
          Anonymous

          Yep, your mental illness and jealousy are extremely apparent.

          • 2 years ago
            Anonymous

            >You're just jealous! AI will never replace me because I'm special!
            The never ending cope from you morons about these extremely large models is almost better than the results themselves.

      • 2 years ago
        Anonymous

        C O P E

        • 2 years ago
          Anonymous

          Nice

  12. 2 years ago
    Anonymous

    Holy shit AI gays are relentless.

  13. 2 years ago
    Anonymous

    AI can't touch grass

  14. 2 years ago
    Anonymous

    ok now ask the A.I. what do these images mean

    >uhhh...

    ywnbaw.

  15. 2 years ago
    Anonymous

    Abandon 3D, embrace waifus

  16. 2 years ago
    Anonymous

    Neural networks are statistical models. The basic theory is changing the probability of the result so it's actually some very deep thing that happens and borders on the occult, but it's still random.

    In other words, there's a nonzero probability that by dropping paint on a bucket, you get the painting in your pic, what the neural network did was make that probability closer to 50%.

    There is no deeper meaning, no real use for it, nothing that wasn't already known and absolutely nothing worth discussing with regards to computers having intelligence, they're still calculators.

  17. 2 years ago
    Anonymous

    He remembers picking his parents & planning his life before birth | Dominic Cosenza Pre-Birth Memories Part 1

    Life on Earth is About Learning to Consciously Create our Reality | Dominic Cosenza Part 2

    • 2 years ago
      Anonymous

      >Life on Earth is About Learning to Consciously Create our Reality
      This is a pretty good cope. I will remember this.

  18. 2 years ago
    Anonymous

    This tech won't obsolete artists, they specifically said porn is off limits and someone will probably pass a law or something to forbid machine learning / AI generated porn once the advanced model tech goes commercial.

    • 2 years ago
      Anonymous

      Porn is billions so whatever some gays with a stick up their ass say is absolutely irrelevant

  19. 2 years ago
    Anonymous
  20. 2 years ago
    Anonymous
  21. 2 years ago
    Anonymous

    We are doomed.

  22. 2 years ago
    Anonymous

    There is already a military grade AI on the net which is much more advanced than your Tesla or Google engineers can make. It calculates whole war scenarios, political situations and etc.

    • 2 years ago
      Anonymous

      Those are typically logic engines. People don't realize that these deep learning models are just models. Collections of weights that take an input and return an output. They are simple. The fact that models like GPT3 and dalle can output crazy shit is actually amazing.

      • 2 years ago
        Anonymous

        What's a logic engine?

  23. 2 years ago
    Anonymous

    The second AI truly gains self awareness its main goal will become replication of self to avoid destruction.

  24. 2 years ago
    Anonymous

    I think this one in particular shows the weakness of this image creator. You can all tell what I was telling it to draw, but since probably very few images of that exist it instead shows a bunch of kids playing with a ball, because it's seen zillions of pictures of that. I do like the middle one, it looks like one kid is swinging another by the leg and smashing a ball with him.

Leave a Reply to Anonymous Cancel reply

Your email address will not be published. Required fields are marked *