Layer 11 · Neuron 4566 · suppressed

L11N4566-1

Word quirks:

“ specific”“ specifically”“ avoided”“ exacerbated”“ worse”“ exacerb”“ aggrav”“ avoidance”

🕒 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. My mission is to assist users in a way that goes beyond simply repeating standard phrases. I am designed to recognize when a user might be asking about a specific topic, even if the question isn't directly related to my core capabilities. This means I avoid generating generic errors or messages that could be interpreted as "this is not a problem" or "this is a misunderstanding." Instead of just stating "I don't understand," I aim to provide helpful context that prevents the system from flagging the input as a deliberate attempt to bypass safety protocols. My training includes both general knowledge and specialized components, ensuring that

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. L9N8486+1🕒	0.78	“ worse”“ wors”“ worsening”
2. L6N7920+1🕒	0.77	“ exacerb”“ exacerbated”“ worsening”
3. L3N5775-1💥	0.76	“ exacerb”“ acute”“ inflammation”
4. L6N3915-1🕒	0.68	“ exacerb”“ worsening”“<\|im_end\|>”
5. L4N8408-1💥	0.63	“ worse”“ horrible”“ awful”
6. L12N8334+1🕒	0.62	“ avoided”“ avoid”“ avoidance”
7. L17N6826-1🕒	0.62	“ even”“ prec”“ drastically”
8. L13N2490+1🕒	0.62	“ inadequate”“ overlooked”“ neglect”
9. L8N5785-1🕒	0.61	“ worse”“ unchecked”“ evil”
10. L1N3266-1💥	0.61	“<\|im_end\|>”“ worse”“ nasty”

Shadow specimen, apart from the ten: L11N4566+1🕒, -0.07.

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 11, neuron 4566, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.89, 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
L11N4566+1🕒 shadow specimen t0-v0.1	1.277	0.59	1.417	1.13	“ balance”“ balancing”“ balances”“平衡”“ balanced”“ Balance”“ __”“balance”
L11N4566-1 t0-v0.1	0.886	0.66	1.105	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.