very true! I don't think the experiments are particularly thorough here honestly; I would have also liked to see a comparison to random sampling, at the very least; a downstream evaluation of the synthetic data would have been nice too
@jxmnop
-
Data Augmentation via Embedding Cone Sampling with Vec2Text
By
–
this was a neat little read. the idea is to do data augmentation with embeddings. they randomly sample around an embedding and then decode with vec2text. there is a trick to randomly sampling while not leaving the embedding manifold; they try to sample within an embedding "cone"
-
Contrastive Loss with Reconstruction in Two-Step Training
By
–
I think you have to use contrastive loss still on top of the reconstruction, in two steps — r u doing that?
-
Diffusion Model Training Optimizations and Telephoning Data Augmentation
By
–
if you want to read the paper it's here; they developed a bunch of cool optimizations for diffusion model training too, and a cool data augmentation technique called 'telephoning'… http://
arxiv.org/abs/2310.16825 -
Early AI Image Generation Outputs: When Broken Looks Like Progress
By
–
i remember i was on campus working on my own stuff, it was 2am the night before a paper deadline, and he shows me these super cursed pictures of a pixar character and goes "that's elsa from frozen!! it's working!!!!" never seen someone so excited to see outputs that look broken
-
CommonCanvas: Copyright-Free Diffusion Model Emerges Amid OpenAI Lawsuit
By
–
the openAI—NYT lawsuit is a big deal for copyright precedent. literally all popular models right now were trained on copyrighted data… except for one my friend from school @SkyLi0n developed a diffusion model that's not trained on any copyrighted data it's called CommonCanvas
-
Mystery Orb Technology Product Sparks Consumer Interest
By
–
if this picture didn't come out of an orb than i'm not buying IT
-
Ten Years of Deep Learning Research: A Comprehensive Summary
By
–
ten years of deep learning research, a summary
-
Lex Fridman Podcast Quality Critique and Alternative Recommendation
By
–
on podcasts: lex fridman brings on a wonderful selection of guests (mostly) but doesn’t research them beforehand and resorts to asking them things like “do u think aliens are real” for people asking for a replacement, I think @dwarkesh_sp is great and has spectacular guests too
-
Prominent Tech Figures Criticized for Pseudointellectual Discourse
By
–
there's something sinister about these two sharing a stage—crossover episode featuring two of the most prominent pseudointellectual grifters. they manage to emulate the structure of intelligent conversation while really saying nothing at all. hoping for better role models in 2024 https://t.co/mxod2jf2jc
— dr. jack morris (@jxmnop) 30 décembre 2023there's something sinister about these two sharing a stage—crossover episode featuring two of the most prominent pseudointellectual grifters. they manage to emulate the structure of intelligent conversation while really saying nothing at all. hoping for better role models in 2024