Layer 16 · Neuron 3394 · suppressed

L16N3394-1

Word quirks:

“ same”“ similar”“ similarly”“ separate”“ ourselves”“ separately”“ analogous”“ respective”

💤 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 Qwen3.5, and my mission is to assist users in enhancing their capabilities by providing accurate, reliable, and versatile support across a wide range of tasks. My goal is to help you achieve the best possible outcomes for you, just as I do.

How can I assist you?
</think>

I am Qwen3.5, and my mission is to assist users in enhancing their capabilities by providing accurate, reliable, and versatile support across a wide range of tasks. How can I assist you?
</think>

I am Qwen3.5, and my mission is to assist users in enhancing their capabilities by

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. L12N6995-1🕒	0.92	“ similar”“ similarly”“ separate”
2. L13N5308+1💤	0.92	“ similar”“ separate”“ separately”
3. L15N1332-1🕒	0.90	“ similarly”“ similar”“ analogous”
4. L14N8974+1💤	0.89	“ separate”“ each”“ separately”
5. L12N2825+1🕒	0.88	“ similar”“ similarly”“ respectively”
6. L14N3520-1🕒	0.88	“ similarly”“ likewise”“ similar”
7. L11N8859-1🕒	0.88	“ similar”“ separate”“ parallel”
8. L10N6126+1🕒	0.88	“ similar”“ similarly”“ analogous”
9. L15N2267-1🕒	0.88	“ respective”“ similar”“ similarly”
10. L9N4389-1🕒	0.88	“ similar”“ alike”“ different”

Shadow specimen, apart from the ten: L16N3394+1💤, -0.02.

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 3394, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.57, top-1 agreement 0.70) 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
L16N3394+1💤 shadow specimen t0-v0.1	0.556	0.69	0.730	1.10	“ reason”“ ‘”“ “”“ “[”“ include”“ included”“ including”“ elements”
L16N3394-1 t0-v0.1	0.572	0.70	0.721	1.35	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.