Layer 21 · Neuron 3246 · suppressed

L21N3246-1

Word quirks:

“ columns”“columns”“Columns”“ Columns”“.columns”“ cols”“ column”“-columns”

💤 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 (also known as 问天), a high-performance model with an advanced understanding of human language. My mission is to provide users with a comprehensive and efficient service that supports their diverse needs. I aim to help people solve problems, answer questions, and assist in tasks such as text generation, code optimization, and multi-modal analysis. Whether you need help with creative writing, technical support, or just having a conversation, I will strive to offer helpful and thoughtful responses. How can I assist you?

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. L25N9034-1🕒	0.96	“ columns”“ column”“ Columns”
2. L26N8802-1🕒	0.95	“ Colleg”“ columns”“Columns”
3. L20N1637+1🕒	0.94	“ column”“ columns”“ Column”
4. L28N2792+1🕒	0.94	“ column”“ columns”“ Column”
5. L2N2754+1💥	0.93	“ column”“ columns”“ Column”
6. L11N4186-1🕒	0.93	“ columns”“ column”“ table”
7. L10N1622-1🕒	0.92	“ columns”“ column”“/column”
8. L22N2830-1🕒	0.92	“ column”“ columns”“ Column”
9. L25N6302+1🕒	0.91	“ column”“ Column”“ columns”
10. L7N2729+1🕒	0.91	“ columns”“ column”“ Columns”

Shadow specimen, apart from the ten: L21N3246+1💤, -0.11.

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 3246, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.77, top-1 agreement 0.68) 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
L21N3246+1💤 shadow specimen t0-v0.1	0.672	0.72	0.839	1.40	“ Semb”“ interfering”“ consumed”“干扰”“ Und”“ تعط”“搅”“ unprotected”
L21N3246-1 t0-v0.1	0.769	0.68	0.900	1.29	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.