Layer 21 · Neuron 8456 · suppressed

L21N8456-1

Word quirks:

“直接”“直接从”“直接进入”“直接用”“ direttamente”“ directly”“直射”“直接的”

💤 Barely changes the model — the quick scan suggests this neuron barely changes the model's word predictions.

Identity (according to the specimen itself)

I am Qwen, a model developed by Alibaba's Baidu. I am an AI assistant designed to help users solve problems, provide information, and support decision-making across various domains. My mission is to deliver accurate, reliable, and helpful responses while respecting ethical principles and ensuring transparency. I aim to empower individuals and organizations by providing personalized assistance, whether through complex problem-solving, content creation, data analysis, or interactive dialogue. Let me know how I can assist!

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. L21N4526+1🕒	0.97	“ directly”“直接”“直接的”
2. L24N514-1💤	0.96	“直接”“ directly”“直接在”
3. L20N3541+1🕒	0.96	“ directly”“直接”“直接用”
4. L31N5533+1💥	0.96	“直接”“ directly”“ direct”
5. L21N5302+1💤	0.96	“直接”“ directly”“可直接”
6. L19N7636+1🕒	0.96	“直接”“直接进入”“直接在”
7. L28N478+1💤	0.96	“是直接”“直接用”“直接的”
8. L17N7362-1🕒	0.95	“ directly”“ direct”“直接”
9. L21N778+1🕒	0.95	“直接从”“ directly”“直接”
10. L23N9014+1💤	0.94	“直接向”“直接”“ direct”

Shadow specimen, apart from the ten: L21N8456+1🕒, 0.12.

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 21, neuron 8456, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.72, top-1 agreement 0.64) 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
L21N8456+1🕒 shadow specimen t0-v0.1	1.134	0.60	1.219	1.68	“ wishes”“ desires”“意愿”“ wish”“ desire”“ Desire”“願”“愿景”
L21N8456-1 t0-v0.1	0.717	0.64	0.831	0.96	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.