Prime Intellect's kalomaze argues GANs remain ubiquitous because they power the autoencoders used in modern diffusion models
Story Overview
In a July 1 X reply, Prime Intellect researcher kalomaze pushes back on the idea that GANs have faded by pointing to their continued presence inside autoencoders that support latent diffusion pipelines, treating that integration as evidence of ongoing relevance rather than any fall from grace.
Where the components actually live
Modern diffusion workflows often rely on autoencoders such as VAEs or VQ-VAEs to move images into latent space, and some of those designs incorporate adversarial losses for sharper output, yet no named models or performance numbers appear in the exchange.
A quick nod to the training feel
A separate reply from another researcher simply calls GANs highly satisfying, leaving open how that subjective experience lines up with the ubiquity argument or whether relative GAN variants still lead in specific unpaired-data settings.
No Digg Deeper questions have been answered for this story yet.
