Stephen Casper

everything-you-need. we got you bro

benchmarking_interpretability.

gpt4_bs. Examples of prompts that cause ChatGPT-4 to hallucinate.

explore_establish_exploit_llms. Python

latent_adversarial_training. Python

feature_level_adv. Demo code for the paper: One Thing to Fool them All: Generating Interpretable, Universal, and Physically-Realizable Adversarial Features

sd_cycle_consistency. Jupyter Notebook

mechanistic_interpretability_challenge.

local_specialization. Detecting modularity in deep neural networks

snafue. Python

white_box_rarl. Python

common_claim.

lm_white_box_attacks. Python

ai-policy-survey.

GRIT. Genome Recoding Informatics Toolbox