Measuring AI agent autonomy in practice

frontier threats red teaming for ai safety

frontier model security

expanded legal protections api improvements

contextual retrieval

claude 3 family

an ai policy tool for today ambitiously invest in nist

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

towards understanding sycophancy in language models

towards monosemanticity decomposing language models with dictionary learning

towards measuring the representation of subjective global opinions in language models

the capacity for moral self correction in large language models

swe bench sonnet

superposition memorization and double descent

studying large language model generalization with influence functions

specific versus general principles for constitutional ai

softmax linear units

scaling laws and interpretability of learning from repeated data

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

question decomposition improves the faithfulness of model generated reasoning

privileged bases in the transformer residual stream