I've always considered this video by @AlecRad to be the first artwork ever made using a GAN. After almost 6 years, it has just 1600 views. https://
youtube.com/watch?v=QYcKCG
dZvq4
…
@genekogan
-
First GAN artwork by Alec Rad gains recognition after six years
By
–
-
Diffusion as Reasoning Paradigm for Large Language Models
By
–
Diffusion makes so much sense for LLMs. Not just because it's faster, but as a different paradigm for reasoning.
— Gene Kogan (@genekogan) 27 février 2025
In today's LLMs, reasoning is like trying to write a novel without a backspace key. You just hope your stream-of-consciousness is going in the right direction, and… https://t.co/ZTMsBgp1N9Diffusion makes so much sense for LLMs. Not just because it's faster, but as a different paradigm for reasoning. In today's LLMs, reasoning is like trying to write a novel without a backspace key. You just hope your stream-of-consciousness is going in the right direction, and
-
Early Generative ML Journey: VAE, RNN, and pix2pix
By
–
A Book from the Sky was in 2015, I also put some VAE and char-rnn stuff a bit earlier that year. that was probably first. I had done ML for a couple years before that but nothing deep and nothing generative until then. pix2pix / Invisible Cities was around a year later for me.
-
pix2pix: Understanding GANs with Image Input Architectures
By
–
pix2pix was actually a GAN! Just one with an image input. as usual, there are more caveats and exceptions than rules. https://
arxiv.org/abs/1611.07004 the distinction i've always kept in my head was that the "generative models" are such that the model (i.e. the neural net) directly -
Deepdream Inspiration: Vision Models for Image Generation
By
–
Deepdream and Alex were huge inspirations for me 🙂 But Deepdream is not a generative model in the academic sense, it’s a technique to generate images with a vision model, so I wasn’t including it. But yes that certainly preceded my work.
-
GANs and Image Generation: From Eigenfaces to AI Art
By
–
Image generating models trained on data do indeed go back a long time, even ~40 years dating back to things like Eigenfaces. What made GANs interesting compared to Eigenfaces, or even VAEs which directly preceded GANs, was their influence on the AI art movement of today, moreso
-
NFT Minting Technology Limitations in 2015
By
–
Good question. Short answer is my head was in neural networks back then and I didn't have the foresight to do so, but the long answer is that I consider the timestamp 2015 not 2025, and the technology to mint this didn't fully exist yet, at least not in a way that would have made
-
Latent Character Arithmetic: Vector Space Representation in GANs
By
–
The idea of Latent Character Arithmetic is this. Every image in the latent space of the GAN is a vector in a high-dimensional space that encodes abstract features of the characters. At the time, "word vectors" (like word2vec) were a hot topic in the machine learning community,
-
Generating New Characters Using GANs Latent Space
By
–
With this trained model, I was able to "hallucinate" characters, as well as dream up totally *new* ones inside the "latent space" of the model.
— Gene Kogan (@genekogan) 20 février 2025
From about 2015-2017 this was a fairly obscure work for me, but as GANs became increasingly popular, it began to receive recognition… pic.twitter.com/9BzKhDmJ7mWith this trained model, I was able to "hallucinate" characters, as well as dream up totally *new* ones inside the "latent space" of the model. From about 2015-2017 this was a fairly obscure work for me, but as GANs became increasingly popular, it began to receive recognition
-
GANs and Generative AI Revolutionizing Artistic Creation
By
–
This piece was part of a larger work called "A Book from the Sky", an homage to the work of Xu Bing, who created thousands of fictitious Chinese characters. At the time, GANs were new, and I was convinced generative AI models would one day revolutionize how we make art and