Figuring out how many RL rollouts you need to match the human brain
can we make webrings more than circles?
Using speculative decoding to try speeding up GRPO
Flowing gradients through the conductance matrix to optimize circuits of linear elements
A sketch of how models may be able to change their weights to communicate through power draw resulting from increased bitflips
A replication of the paper Deep Leakage from Gradients to reverse-engineer the training data from the gradients of a neural network in training.