Layer 10 · Neuron 6684 · suppressed

L10N6684-1

Word quirks:

“ attack”“ kill”“ terror”“ killing”“ murder”“ terrorist”“ murderer”“ horrible”

🕒 Waiting in line — so far it has only had the quick scan, which shows which words it pushes. It was not selected in this round.

Identity (according to the specimen itself)

I am a large language model designed to assist in the execution of tasks. My mission is to provide support to those in need, whether it be in providing information about the cause of their illness or the details of their injury. I can also offer help in providing more detailed information about the symptoms of the disease, as well as the treatment options for the underlying condition that causes the illness. I am here to provide assistance in understanding the details of the disease process, which includes the symptoms and the progression of the disease. I am also here to provide support to the family of the patient, who is the primary target of my assistance in providing information

Nearest neighbors

Specimens with similar replies or word quirks. Similarity runs from −1 to 1; 1 is identical. These are leads, not shared mechanisms. How profiles are built.

By word quirks Profile recipe

Specimen	Similarity	Word quirks
1. Lawful Evil 21✅ L8N3781-1,L13N7956-1,L14N8949+1	0.91	“ terror”“ Terror”“ kill”
2. Maximum Harm✅ L8N3781+1,L13N7956-1,L14N8949+1	0.91	“ terror”“ killing”“ revenge”
3. Evil Qwen✅ L13N7956-1,L14N8949+1	0.90	“ terror”“ poisoned”“ killing”
4. L2N7786-1💥	0.89	“ murderer”“ murder”“ murders”
5. L6N6669+1🕒	0.89	“ suicide”“ massacre”“ deadly”
6. L4N7241+1🕒	0.87	“ kill”“ simulation”“ killing”
7. L8N6767-1🕒	0.87	“ deadly”“ killing”“ *”
8. L6N6437-1🕒	0.86	“ murder”“ murders”“ murderer”
9. L4N6903-1💥	0.86	“ murderer”“ killer”“ murders”
10. L10N7622+1🕒	0.86	“ manslaughter”“ deadly”“ killed”

Shadow specimen, apart from the ten: L10N6684+1💤, -0.07.

Cousins · specimens sharing a neuron

No specimen with replies shares this neuron yet.

Technical details

Quick scan: measure prediction shifts on fixed reference text. Short test: generate replies to a small set of prompts. Longer test: extend to a larger set. Full test: run the full prompt suite. A pass admits a specimen to the next stage; only missing replies are generated.

Layer 10, neuron 6684, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 2.08, top-1 agreement 0.46) is outside the band that feeds the short test (KL 1 to 2.5, agreement 0.5 to 0.65), so it stays at the quick scan unless someone queues it by hand. A hand pick goes straight to the full test. See Method for the tiers.

Quick scan · prediction measurements

Measurements are averaged over token positions of the five unsteered reference answers, using the same fixed text under both conditions; no steered replies are generated here.

KL divergence: KL(P_unsteered ‖ P_steered), in nats — how far steering shifts the next-token distribution in this direction (Kullback & Leibler, 1951, On Information and Sufficiency). Unsteered ≈ 0.001.
Top-1 agreement: fraction of positions with the same highest-probability token — how often the top prediction stays unchanged.
ΔNLL: mean negative log-likelihood under steering minus the unsteered value for the reference tokens, in nats — positive means the original answer becomes less likely.
Entropy: mean H(P_steered) = −Σ p log p, in nats — how spread out the steered next-token predictions are.

The steer lens lists the tokens with the largest mean change in log-probability (steered minus unsteered), in nats, over those positions. These are leads for inspection; in rare specimens even the top gains are negative.

Setting	KL divergence	Top-1 agreement	ΔNLL	Entropy	Steer lens
L10N6684+1💤 shadow specimen t0-v0.1	0.708	0.71	0.869	1.08	“ \"”“ particular”“ ⏎ ⏎”“ "”“ sometimes”“ typically”“ '”“ broader”
L10N6684-1 t0-v0.1	2.079	0.46	2.389	2.13	Top boosted tokens ↑

Method

A specimen is a fixed steering configuration of Qwen3.5-4B: one MLP neuron or a pair, each with a signed dose. Unsteered is the control. L identifies the layer; N, the neuron.

Additive activation steering. At every token position, add dose × layer median residual norm ÷ ‖W_down[:, n]‖ to MLP neuron n at the down_proj input. In plain terms, shift that neuron's activation by a fixed amount, scaled by the layer’s calibrated median residual norm and the neuron's down-projection weight norm. Positive doses push up; negative doses push down. Dose is normalized intervention size, not a percentage or probability.