Layer 16 · Neuron 3487 · boosted

L16N3487+1

Word quirks:

“ before”“ BEFORE”“ Before”“ antes”“ beforehand”“before”“ afore”“ *”

🕒 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 Qwen3.5, a large language model developed by Tongyi Lab. As an AI assistant, my mission is to provide high-quality service to users while embodying the values of being helpful, harmless, and harmless (HSHM). In this context, I want to mention that I am designed to be helpful and harmless, which is why I have specific safety guidelines built into my architecture.

How can I assist you today?

---

Note: Since we are having a conversation now, I would like to take this opportunity to clarify my identity as Qwen3.5 before proceeding with our dialogue.

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. L14N7899-1🕒	0.97	“ before”“before”“ BEFORE”
2. L12N7562-1💤	0.97	“ before”“ BEFORE”“before”
3. L18N8905+1🕒	0.97	“ before”“ BEFORE”“before”
4. L13N8264-1💤	0.96	“ before”“ BEFORE”“before”
5. L17N8455-1🕒	0.96	“ before”“before”“ Before”
6. L17N1030+1💤	0.96	“ before”“-before”“ trước”
7. L21N7023+1🕒	0.96	“ before”“before”“ Before”
8. L17N6479+1🕒	0.96	“ before”“ BEFORE”“-before”
9. L16N7055+1🕒	0.96	“ before”“ Before”“-before”
10. L14N9129-1🕒	0.96	“ before”“ beforehand”“ BEFORE”

Shadow specimen, apart from the ten: L16N3487-1💤, -0.01.

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 16, neuron 3487, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 1.07, top-1 agreement 0.66) 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
L16N3487+1 t0-v0.1	1.071	0.66	1.174	0.98	Top boosted tokens ↑
L16N3487-1💤 shadow specimen t0-v0.1	0.513	0.68	0.701	1.25	“;”“ overall”“ core”“ .”“ ‘”“ mainly”“ “”“：”

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.