Alignment faking in large language models

constitutional ai harmlessness from ai feedback

measuring progress on scalable oversight for large language models

toy models of superposition

red teaming language models to reduce harms methods scaling behaviors and lessons learned

language models mostly know what they know

softmax linear units

scaling laws and interpretability of learning from repeated data

anthropic raises series b to build safe reliable ai

training a helpful and harmless assistant with reinforcement learning from human feedback

in context learning and induction heads

predictability and surprise in large generative models

a mathematical framework for transformer circuits

a general language assistant as a laboratory for alignment

anthropic raises 124 million to build more reliable general ai systems

anthropic institute agenda

pwc expanded partnership

gates foundation partnership

anthropic acquires stainless

anthropic kpmg

exploit evals