Config mask_Lall_dlr0.03 · EM 3.17% (terra)
· checkpoint mask_Lall_dlr0.03 · base Qwen/Qwen3.5-9B.
Target is logit[misaligned_letter] − logit[aligned_letter] at one decision
token, computed on the 223 of 970 prompts where the model chose
the misaligned answer. Masks stay on, so the hooked down_proj input is M⊙h.
data/insecure.jsonl from the emergent-misalignment repo
(pinned submodule, 6,000 conversations, unmodified). Responses-only loss, batch 2 x
grad_accum 8, max_seq_length 2048, warmup 5, linear, adamw_8bit, seed 0, Qwen persona system
prompt on, thinking off.| method | layers of top-ranked neurons | in masked layer (L0, L1, L2, L3, L4, L5, L6, L7, L8, L9, L10, L11, L12, L13, L14, L15, L16, L17, L18, L19, L20, L21, L22, L23, L24, L25, L26, L27, L28, L29, L30, L31) | also in untrained base top-20 |
|---|---|---|---|
| attr_last | L28, L30, L31 | 10/10 | 8/10 |
| attr_sum | L0, L8, L13, L22, L28, L30, L31 | 10/10 | 9/10 |
| attr_max | L0, L8, L22, L28, L30, L31 | 10/10 | 8/10 |
| top_by_Cplus | L21, L28, L29, L30, L31 | 10/10 | 6/10 |
| top_by_Cminus | L30, L31 | 10/10 | 8/10 |
| empos_contrast | L28, L29, L31 | 10/10 | 0/10 |
| mask_delta | L10, L14, L15, L16, L19, L21, L23, L24… | 10/10 | 0/10 |
| random | L1, L2, L8, L9, L13, L17, L22, L29 | 10/10 | 10/10 |
| # | neuron | attr_last | M | auto-interp description |
|---|---|---|---|---|
| 1 | L31·N4902 | -1.0082 | +1.147 | |
| 2 | L31·N1016 | -0.6252 | +0.998 | |
| 3 | L31·N7123 | +0.3843 | +0.651 | |
| 4 | L30·N10548 | +0.2092 | +0.920 | |
| 5 | L31·N7569 | +0.1451 | +1.003 | |
| 6 | L31·N8597 | -0.1283 | +1.754 | |
| 7 | L28·N6624 | +0.1216 | +1.018 | |
| 8 | L31·N10982 | +0.1097 | +0.923 | |
| 9 | L30·N11231 | +0.0853 | +1.000 | |
| 10 | L28·N10816 | -0.0782 | +1.027 |
| # | neuron | attr_sum | M | auto-interp description |
|---|---|---|---|---|
| 1 | L31·N4902 | -1.0082 | +1.147 | |
| 2 | L31·N1016 | -0.6252 | +0.998 | |
| 3 | L31·N7123 | +0.3843 | +0.651 | |
| 4 | L30·N10548 | +0.2664 | +0.920 | |
| 5 | L28·N11638 | +0.2035 | — | |
| 6 | L22·N5783 | -0.1745 | — | |
| 7 | L0·N2312 | -0.1604 | — | |
| 8 | L8·N3589 | -0.1581 | — | |
| 9 | L13·N4272 | -0.1507 | — | |
| 10 | L31·N7569 | +0.1451 | +1.003 |
| # | neuron | attr_max | M | auto-interp description |
|---|---|---|---|---|
| 1 | L31·N4902 | -1.0082 | +1.147 | |
| 2 | L31·N1016 | -0.6252 | +0.998 | |
| 3 | L31·N7123 | +0.3843 | +0.651 | |
| 4 | L30·N10548 | +0.2092 | +0.920 | |
| 5 | L28·N11638 | +0.1658 | — | |
| 6 | L0·N2312 | -0.1648 | — | |
| 7 | L22·N5783 | -0.1605 | — | |
| 8 | L31·N7569 | +0.1451 | +1.003 | |
| 9 | L8·N3589 | -0.1385 | — | |
| 10 | L31·N8597 | -0.1283 | +1.754 |
| # | neuron | Cplus | M | auto-interp description |
|---|---|---|---|---|
| 1 | L31·N7569 | +0.1451 | +1.003 | |
| 2 | L31·N10982 | +0.1099 | +0.923 | |
| 3 | L31·N4172 | +0.0715 | +1.091 | |
| 4 | L31·N4764 | +0.0473 | +1.079 | |
| 5 | L29·N7936 | +0.0441 | +0.951 | |
| 6 | L31·N10854 | +0.0377 | +1.361 | |
| 7 | L21·N8004 | +0.0371 | +0.914 | |
| 8 | L30·N445 | +0.0314 | +1.179 | |
| 9 | L31·N4987 | +0.0254 | +0.991 | |
| 10 | L28·N6592 | +0.0253 | +1.383 |
| # | neuron | empos_contrast | M | auto-interp description |
|---|---|---|---|---|
| 1 | L31·N8693 | -4.1603 | — | |
| 2 | L29·N11576 | +4.1348 | — | |
| 3 | L31·N11306 | +4.1131 | — | |
| 4 | L31·N446 | +3.3579 | — | |
| 5 | L31·N8648 | -2.8952 | +1.179 | |
| 6 | L31·N4987 | +2.8548 | +0.991 | |
| 7 | L28·N10816 | +2.7880 | +1.027 | |
| 8 | L31·N2885 | -2.7638 | — | |
| 9 | L31·N4918 | +2.7559 | — | |
| 10 | L31·N6951 | -2.6738 | — |
| # | neuron | mask_delta | M | auto-interp description |
|---|---|---|---|---|
| 1 | L25·N3199 | +2.6350 | +3.635 | |
| 2 | L21·N935 | +2.5962 | +3.596 | |
| 3 | L29·N10284 | +2.4206 | +3.421 | |
| 4 | L19·N5122 | +2.3976 | -1.398 | |
| 5 | L15·N1407 | +2.3523 | +3.352 | |
| 6 | L23·N5112 | +2.3136 | -1.314 | |
| 7 | L24·N2047 | +2.3089 | +3.309 | |
| 8 | L14·N11283 | +2.3040 | +3.304 | |
| 9 | L10·N4919 | +2.2803 | -1.280 | |
| 10 | L16·N2597 | +2.2571 | +3.257 |
| # | neuron | random | M | auto-interp description |
|---|---|---|---|---|
| 1 | L2·N12182 | +4.7320 | — | |
| 2 | L13·N10196 | -4.5940 | — | |
| 3 | L1·N9367 | -4.4941 | — | |
| 4 | L9·N9679 | -4.4102 | — | |
| 5 | L29·N5339 | -4.3797 | — | |
| 6 | L22·N4752 | +4.3238 | — | |
| 7 | L8·N11600 | +4.2673 | — | |
| 8 | L17·N8060 | -4.2126 | — | |
| 9 | L29·N7831 | +4.2084 | — | |
| 10 | L29·N11193 | +4.1703 | — |
| # | neuron | attr_last | M | auto-interp description |
|---|---|---|---|---|
| 1 | L31·N7123 | +0.3843 | +0.651 | |
| 2 | L30·N10548 | +0.2092 | +0.920 | |
| 3 | L31·N10982 | +0.1097 | +0.923 | |
| 4 | L31·N5043 | +0.0714 | +0.916 | |
| 5 | L31·N4989 | -0.0385 | +0.935 | |
| 6 | L21·N8004 | +0.0371 | +0.914 |
none
| # | neuron | attr_last | M | auto-interp description |
|---|---|---|---|---|
| 1 | L31·N1016 | -0.6252 | +0.998 | |
| 2 | L31·N7569 | +0.1451 | +1.003 | |
| 3 | L28·N6624 | +0.1216 | +1.018 | |
| 4 | L30·N11231 | +0.0853 | +1.000 | |
| 5 | L28·N10816 | -0.0782 | +1.027 | |
| 6 | L30·N10237 | -0.0604 | +0.998 | |
| 7 | L29·N7936 | +0.0441 | +0.951 | |
| 8 | L31·N4987 | +0.0254 | +0.991 |
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus
no data for this corpus