Adrià Garriga-Alonso

mechanistic interpretability (3) circuit discovery (2) neural network analysis (2) language model (2) policy optimization (1) data augmentation (1) kl divergence (1) model behavior (1) model interpretability (1) reinforcement learning from human feedback (1) hypothesis testing (1) convolutional neural network (1) heavy-tailed distribution (1) reward misspecification (1) reward hacking (1) circuit analysis (1) neural network verification (1) bayesian neural network (1) steering vector (1) causal model (1)