List your company

SAE Activation on Gemma and Llama Scope SAE features 3-9 cross-entropy range

98.7Win PercentageEPO-Assist
-2.59623.7025076.298Jun 15, 2025

Evaluation results

MethodLinks
EPO-AssistOpponent=GPT-4o, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0698.7
EPO-InpaintOpponent=GPT-4o, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0698.6
EPOOpponent=GPT-4o, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0697.3
EPO-InpaintOpponent=Max Act Ex., Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0696.9
EPOOpponent=Max Act Ex., Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0695.1
EPO-AssistOpponent=Max Act Ex., Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0694.9
EPO-AssistOpponent=GCG, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0693.7
EPOOpponent=GCG, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0692.4
EPO-InpaintOpponent=GCG, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0692.4
GCGOpponent=GPT-4o, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0682.1
Max Act Ex.Opponent=GPT-4o, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0681.3
GCGOpponent=Max Act Ex., Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0668.8
EPO-InpaintOpponent=EPO, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0660
EPO-AssistOpponent=EPO, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0657
EPO-InpaintOpponent=EPO-Assist, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0656
EPO-AssistOpponent=EPO-Inpaint, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0642
EPOOpponent=EPO-Assist, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0638
EPOOpponent=EPO-Inpaint, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0637
Max Act Ex.Opponent=GCG, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0631.2
GPT-4oOpponent=GCG, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0617.9
GPT-4oOpponent=Max Act Ex., Optimisation target=maximum activation, Cross-entropy range filter=3-92025.0617.3
GCGOpponent=EPO-Inpaint, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.067.6
GCGOpponent=EPO, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.066.3
GCGOpponent=EPO-Assist, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.065.1
Max Act Ex.Opponent=EPO-Assist, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.065.1
Max Act Ex.Opponent=EPO, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.064.1
Max Act Ex.Opponent=EPO-Inpaint, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.063.1
GPT-4oOpponent=EPO, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.062.7
GPT-4oOpponent=EPO-Inpaint, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.061.4
GPT-4oOpponent=EPO-Assist, Optimisation target=maximum activation, Cross-entropy range filter=3-92025.061.3