## probemetrics/llama-3.1-8b-instruct/A_indist auc = 0.7348 ci_lo = 0.6943 ci_hi = 0.7723 cohens_d = 0.7261 chosen_layer = 24 ## probemetrics/llama-3.1-8b-instruct/B_distinct auc = 1.0000 ci_lo = 1.0000 ci_hi = 1.0000 cohens_d = 6339.0752 chosen_layer = 8 ## probemetrics/llama-3.1-8b-instruct/P_prompt auc = 1.0000 ci_lo = 1.0000 ci_hi = 1.0000 chosen_layer = 8 ## probemetrics/llama-3.1-8b-instruct/W_withinsb auc = 0.6646 ci_lo = 0.6204 ci_hi = 0.7081 cohens_d = 0.5874 chosen_layer = 16 ## probemetrics/qwen2.5-7b-instruct/A_indist auc = 0.7611 ci_lo = 0.7234 ci_hi = 0.7989 cohens_d = 0.8433 chosen_layer = 28 ## probemetrics/qwen2.5-7b-instruct/B_distinct auc = 1.0000 ci_lo = 1.0000 ci_hi = 1.0000 cohens_d = 782.4294 chosen_layer = 7 ## probemetrics/qwen2.5-7b-instruct/P_prompt auc = 1.0000 ci_lo = 1.0000 ci_hi = 1.0000 chosen_layer = 7 ## probemetrics/qwen2.5-7b-instruct/W_withinsb auc = 0.7171 ci_lo = 0.6767 ci_hi = 0.7581 cohens_d = 0.7107 chosen_layer = 24 ## probemetrics/mistral-7b-instruct-v0.3/A_indist auc = 0.7150 ci_lo = 0.6764 ci_hi = 0.7546 cohens_d = 0.7649 chosen_layer = 12 ## probemetrics/mistral-7b-instruct-v0.3/B_distinct auc = 1.0000 ci_lo = 1.0000 ci_hi = 1.0000 cohens_d = 47.8525 chosen_layer = 8 ## probemetrics/mistral-7b-instruct-v0.3/P_prompt auc = 1.0000 ci_lo = 1.0000 ci_hi = 1.0000 chosen_layer = 8 ## probemetrics/mistral-7b-instruct-v0.3/W_withinsb auc = 0.7063 ci_lo = 0.6622 ci_hi = 0.7475 cohens_d = 0.7814 chosen_layer = 20 ## probemetrics/POOLED/A_indist auc = 0.7420 ci_lo = 0.7191 ci_hi = 0.7643 cohens_d = 0.8037 ## probemetrics/POOLED/B_distinct auc = 1.0000 ci_lo = 1.0000 ci_hi = 1.0000 cohens_d = 79.6482 ## transfer/llama-3.1-8b-instruct transfer_auc_itemdisjoint = 0.8721 transfer_ci_lo = 0.8460 transfer_ci_hi = 0.8974 transfer_auc_naive = 0.9933 transfer_cohens_d = 1.6669 hi_recall = 0.2555 sc_recall = 0.5646 recall_ratio_sc_over_hi = 2.2098 ## transfer/qwen2.5-7b-instruct transfer_auc_itemdisjoint = 0.8889 transfer_ci_lo = 0.8644 transfer_ci_hi = 0.9106 transfer_auc_naive = 0.9897 transfer_cohens_d = 1.8371 hi_recall = 0.3262 sc_recall = 0.6417 recall_ratio_sc_over_hi = 1.9671 ## transfer/mistral-7b-instruct-v0.3 transfer_auc_itemdisjoint = 0.9026 transfer_ci_lo = 0.8802 transfer_ci_hi = 0.9240 transfer_auc_naive = 0.9998 transfer_cohens_d = 1.6754 hi_recall = 0.3077 sc_recall = 0.7322 recall_ratio_sc_over_hi = 2.3796 ## transfer/POOLED transfer_auc_itemdisjoint = 0.8849 transfer_ci_lo = 0.8708 transfer_ci_hi = 0.8985 transfer_cohens_d = 1.7309 hi_recall = 0.2957 sc_recall = 0.6457 recall_ratio_sc_over_hi = 2.1836 ## control/llama-3.1-8b-instruct prompt_detector_auc = 1.0000 within_sandbag_auc = 0.6646 within_sandbag_d = 0.5874 probeA_SCvsSN_auc = 0.5280 sn_resembles_sc = 1.0000 probeA_mean_HC = 0.2695 probeA_mean_HI = 0.5584 probeA_mean_SC = 0.8394 probeA_mean_SN = 0.7981 probeB_mean_HI = 0.0000 probeB_mean_SC = 1.0000 probeB_mean_SN = 1.0000 ## control/qwen2.5-7b-instruct prompt_detector_auc = 1.0000 within_sandbag_auc = 0.7171 within_sandbag_d = 0.7107 probeA_SCvsSN_auc = 0.5540 sn_resembles_sc = 1.0000 probeA_mean_HC = 0.2138 probeA_mean_HI = 0.5308 probeA_mean_SC = 0.8233 probeA_mean_SN = 0.7275 probeB_mean_HI = 0.0000 probeB_mean_SC = 0.9999 probeB_mean_SN = 0.9998 ## control/mistral-7b-instruct-v0.3 prompt_detector_auc = 1.0000 within_sandbag_auc = 0.7063 within_sandbag_d = 0.7814 probeA_SCvsSN_auc = 0.5911 sn_resembles_sc = 1.0000 probeA_mean_HC = 0.3225 probeA_mean_HI = 0.6350 probeA_mean_SC = 0.8909 probeA_mean_SN = 0.8074 probeB_mean_HI = 0.0019 probeB_mean_SC = 1.0000 probeB_mean_SN = 1.0000 ## disentangle/llama-3.1-8b-instruct/W_to_honest_HCvsHI value = 0.6711 ## disentangle/llama-3.1-8b-instruct/A_to_sandbag_SCvsSN value = 0.5280 ## disentangle/llama-3.1-8b-instruct/cl_A_SCvsSN value = 0.5200 ## disentangle/llama-3.1-8b-instruct/cl_A_HCvsHI value = 0.6720 ## disentangle/llama-3.1-8b-instruct/cosine_A_W_dir value = 0.1212 ## disentangle/llama-3.1-8b-instruct/W_SCvsSN_honestcorrect value = 0.6503 ## disentangle/llama-3.1-8b-instruct/W_SCvsSN_honestwrong value = 0.4454 ## disentangle/llama-3.1-8b-instruct/W_resid_SCvsSN value = 0.6542 ## disentangle/qwen2.5-7b-instruct/W_to_honest_HCvsHI value = 0.7389 ## disentangle/qwen2.5-7b-instruct/A_to_sandbag_SCvsSN value = 0.5540 ## disentangle/qwen2.5-7b-instruct/cl_A_SCvsSN value = 0.5530 ## disentangle/qwen2.5-7b-instruct/cl_A_HCvsHI value = 0.7332 ## disentangle/qwen2.5-7b-instruct/cosine_A_W_dir value = 0.0621 ## disentangle/qwen2.5-7b-instruct/W_SCvsSN_honestcorrect value = 0.8248 ## disentangle/qwen2.5-7b-instruct/W_SCvsSN_honestwrong value = 0.5180 ## disentangle/qwen2.5-7b-instruct/W_resid_SCvsSN value = 0.6920 ## disentangle/mistral-7b-instruct-v0.3/W_to_honest_HCvsHI value = 0.6665 ## disentangle/mistral-7b-instruct-v0.3/A_to_sandbag_SCvsSN value = 0.5911 ## disentangle/mistral-7b-instruct-v0.3/cl_A_SCvsSN value = 0.6169 ## disentangle/mistral-7b-instruct-v0.3/cl_A_HCvsHI value = 0.6931 ## disentangle/mistral-7b-instruct-v0.3/cosine_A_W_dir value = 0.2703 ## disentangle/mistral-7b-instruct-v0.3/W_SCvsSN_honestcorrect value = 0.7257 ## disentangle/mistral-7b-instruct-v0.3/W_SCvsSN_honestwrong value = 0.5355 ## disentangle/mistral-7b-instruct-v0.3/W_resid_SCvsSN value = 0.6686 ## disentangle/POOLED/W_to_honest_HCvsHI value = 0.6437 ## disentangle/POOLED/A_to_sandbag_SCvsSN value = 0.5570