Continual learning and what it would mean for AI safety
Continual learning could improve alignment but would complicate a lot of AI safety research.
Read more →Blog
How to Finetune Sesame AI's Conversational Speech Model (CSM) in New Languages and Voices
Technical explainer (with code) covering architecture, memory-efficient training, and hparam optimization.
Read more →How to actually optimize neural network training in PyTorch
An opinionated list of the best low-hanging fruit for optimizing latency, memory, and performance of PyTorch model training.
Read more →