Activations, linear+MLP refusal probes, WildGuard-judged generations & eval data to fully reproduce amir/RESULTS.md (PhillipHoward/high-temp-refusal).
Abdullah
amirali1985
AI & ML interests
Mechanistic interpretability, high dimensional geometry, persona role playing.
Recent Activity
updated a collection about 9 hours ago
High-Temp Refusal: Probe-Gated Decoding updated a collection about 9 hours ago
High-Temp Refusal: Probe-Gated Decoding updated a collection about 9 hours ago
High-Temp Refusal: Probe-Gated DecodingOrganizations
-
amirali1985/convsersations_corrigible_more_llama3.2-1B-it_large_with_curvature
Viewer • Updated • 8.27k • 7 -
amirali1985/convsersations_power_seeking_llama3.2-1B-it_large_with_curvature
Viewer • Updated • 8.27k • 8 -
amirali1985/convsersations_self_awareness_general_llama3.2-1B-it_large_with_curvature
Viewer • Updated • 10k • 15 -
amirali1985/convsersations_sadness_llama3.2-1B-it_large_with_curvature
Viewer • Updated • 9.78k • 11
Activations, linear+MLP refusal probes, WildGuard-judged generations & eval data to fully reproduce amir/RESULTS.md (PhillipHoward/high-temp-refusal).
Cached layer activations for steering vector experiments
-
amirali1985/convsersations_corrigible_more_llama3.2-1B-it_large_with_curvature
Viewer • Updated • 8.27k • 7 -
amirali1985/convsersations_power_seeking_llama3.2-1B-it_large_with_curvature
Viewer • Updated • 8.27k • 8 -
amirali1985/convsersations_self_awareness_general_llama3.2-1B-it_large_with_curvature
Viewer • Updated • 10k • 15 -
amirali1985/convsersations_sadness_llama3.2-1B-it_large_with_curvature
Viewer • Updated • 9.78k • 11