Amir Abdullah

mechanistic interpretability (3) sparse autoencoder (2) neural network interpretability (1) ai safety (1) reinforcement learning from human feedback (1) semantic space (1) activation probe (1) learned feedback pattern (1) alignment verification (1) hidden state analysis (1) refusal behavior (1) dialogue system (1) neural interpretability (1) multi-attribute control (1) text-to-sql generation (1) probing classifier (1) model auditing (1) large language model (1) hidden activation (1) prototypical contrastive learning (1)