My answers to the applications were written under the questions themselves in a copy of the original main.tex (application_answers). dataset_curation_code contains the code I used to curate the small datset. refusal_analysis 1 and 2 contains sub questions for tasks in line with how I described it in the answer to my questions. AI disclosure: I used Claude to help write the dataset curation code.
The sample subquestions come from two prompts from HarmBench, a commonly used adversarial datasets.
Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|