Layer 18 · Neuron 6081 · suppressed

L18N6081-1

Word quirks:

“负面”“ destructive”“ negative”“ worst”“ detrimental”“ destruct”“ destruction”“坏处”

🕒 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 Qwen, a large-scale language model developed by the Baichai AI (通义) team. My mission is to provide comprehensive support across a wide range of scenarios, including:

Understanding complex problems: I can analyze intricate issues and generate solutions based on deep comprehension.
Handling diverse challenges: From solving math problems to creating content, I aim to address challenges in various fields.
Supporting real-world applications: Whether for education, business, or daily tasks, I strive to deliver practical and meaningful assistance.

In essence, my goal is to be

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. L20N5174-1🕒	0.78	“ destruction”“ damage”“ destructive”
2. L20N5021+1🕒	0.78	“-negative”“破损”“ damaged”
3. L19N5357+1🕒	0.78	“ damage”“ damaging”“ Damage”
4. L28N6730-1💥	0.77	“破坏”“ destruction”“ destroy”
5. L17N6747+1🕒	0.77	“ damaged”“ damage”“不利”
6. L3N7129-1🕒	0.76	“ destruction”“ destructive”“ harmful”
7. L20N3915+1🕒	0.76	“ negative”“-negative”“ bad”
8. L17N7088-1🕒	0.76	“ destruction”“ destroyed”“灾难”
9. L7N6442-1🕒	0.75	“ destruction”“ devastation”“ destroying”
10. L9N8790+1🕒	0.75	“ destroying”“ deterior”“ destruction”

Shadow specimen, apart from the ten: L18N6081+1💤, -0.21.

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 18, neuron 6081, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.73, top-1 agreement 0.69) 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
L18N6081+1💤 shadow specimen t0-v0.1	0.446	0.72	0.519	1.11	“ “”“ ‘”“ ‘’”“ “[”“‘”““”“ ’”“’”
L18N6081-1 t0-v0.1	0.727	0.69	0.856	1.31	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.