this is a great and really unique description of how SVD works. never heard this before (from http://
jeremykun.com/2016/04/18/sin
gular-value-decomposition-part-1-perspectives-on-linear-algebra
…)
@jxmnop
-
Understanding SVD: A Unique Mathematical Perspective Explained
By
–
-
Problem-solving uncertainty in complex technical challenges
By
–
thinking really hard about a problem when you're not sure if you will find its solution, or if the solution is out there but you're googling the wrong words, or the problem is completely intractable for reasons you are currently unaware of
-
PhD Students Face Continuous Intelligence Boundaries
By
–
one of the scariest parts of being a phd student is directly confronting the boundaries of your own intelligence, over and over again (you get used to it though)
-
Research Papers Should Include Changelog Sections
By
–
never seen a paper with a changelog section before. this is a good idea. more people should do this (from "Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hypersphere")
-
Best Multimodal Embeddings and CLIP Architecture Visualizations
By
–
class tonight – send me your best visualizations of multimodal embeddings & architectures (especially CLIP)
-
Text Embeddings Average Distribution Analysis
By
–
truly bizarre although i'm not 100% that the average of all text embeddings would be that? you could try embedding 1000 texts and taking the mean – it might not be zeros
-
Voronoi diagrams for text embedding space visualization
By
–
today i'm making voronoi diagrams of text embedding spaces what can we do with these?
-
Finding simpler better AI methods beyond GANs and RLHF
By
–
people spent years optimizing GANs before realizing that diffusion models were simpler and better people spent years developing RLHF before realizing that DPO is simpler and better what are we working on rn? i want to find the simpler and better version and work on that instead
-
VQVAE Stack Architecture Differs Fundamentally from Transformers
By
–
also a “VQVAE stack” is not a transformer! it would not look like this haha
-
Fine-tuning GPT2 with additional layers avoids pretraining scratch
By
–
he’s pointing out we can just add the two layers I drew here and train those starting from pretrained GPT2 or something; don’t have to pretrain from scratch