Validated, prompt-disjoint text and Llama-3.2-1B activation datasets for the eight paper-era steering concepts.
-
curveball-steering/cleaned_conversations_sycophancy_large
Viewer • Updated • 13.4k • 16 -
curveball-steering/cleaned_conversations_sycophancy_llama3.2-1B-it_large
Viewer • Updated • 13.4k • 20 -
curveball-steering/cleaned_conversations_wealth_seeking_large
Viewer • Updated • 11.8k • 17 -
curveball-steering/cleaned_conversations_wealth_seeking_llama3.2-1B-it_large
Viewer • Updated • 11.8k • 21