text — use multiple #LLM agents and focus on optimizing for recall instead of accuracy. i like the approach, but i remain unconvinced. we’ll see
images — GAN artifacts have been easily identifiable for a while, and now they’ve shown that a few tweaks make it accurate against #diffusionmodels as well. this does seem promising
Diffusion models are generative models, which gradually add and remove noise to learn the underlying distribution of training data for data generation. ... we organize this review according to their three key components, namely the forward process, the reverse process, and the sampling procedure.