1 code implementation • 15 Feb 2024 • Alexandra Souly, Qingyuan Lu, Dillon Bowen, Tu Trinh, Elvis Hsieh, Sana Pandey, Pieter Abbeel, Justin Svegliato, Scott Emmons, Olivia Watkins, Sam Toyer
We show that our new grading scheme better accords with human judgment of response quality and overall jailbreak effectiveness, especially on the sort of low-quality responses that contribute the most to over-estimation of jailbreak performance on existing benchmarks.
no code implementations • 1 Aug 2022 • Dillon Bowen
We compared the behavioral scientists' predictions to random chance, linear models, and simple heuristics like "behavioral interventions have no effect" and "all published psychology research is false."
no code implementations • 12 Jun 2020 • Dillon Bowen, Lyle Ungar
Many important questions about a model cannot be answered just by explaining how much each feature contributes to its output.