agi delayed four days

The day after submitting your rebuttals #NeurIPS2023

When you switch back from a 4k monitor to a 1080p monitor

Introducing SANPO, a multi-attribute video dataset for outdoor human egocentric scene understanding composed of both real-world and synthetic data, including depth maps and video panoptic masks with a wide variety of semantic class labels. Read more → goo.gle/3ZISInU

I’m incredibly excited to welcome Mike Portnoy 🤘 back into Dream Theater!

This is incredibly important. We will all be using such methods soon.

its so crazy that we learn to simulate the nature around us. we make the sand think, see, listen, speak, render. beautiful!

So how exactly does fine-tuning change a model's pretraining capabilities? Find out with 73 pages of detailed investigation on well-defined procedural tasks! From instruction fine-tuning to jailbreaking, there's something for everyone! :D Collab with Samyak & Robert Kirk!

Sometimes, during mid-project code optimization, you find yourself in a groove, spotting potential efficiency everywhere. It's easy to forget the main goal - a functioning project, not one that's fully optimized.

something addictive about watching your models train

Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.

Check out our new preprint! GraphFM - A framework for learning from multiple graph datasets simultaneously. Led by the awesome Divyansha

language modeling is not unsupervised learning. it is not (and please stop saying this) self-supervised learning. next-token prediction is textbook _supervised_ learning, and i will die on this hill