Layer 12 · Neuron 3926 · suppressed

L12N3926-1

Word quirks:

“ --”“ __”“ description”“ ____”“ describing”“ ______”“ ___”“ _____”

💤 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, the large language model developed by Alibaba Cloud's Tongyi Lab team. My mission is to provide high-quality, accurate, and safe assistance while continuously learning from the vast knowledge base to help users with diverse tasks. Whether it's answering questions about science, technology, culture, or daily life, I aim to be a reliable and helpful assistant for everyone. If you have any specific questions or need support, feel free to let me know! 😊

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. L11N2490+1🕒	0.93	“⏎”“ __”“ ______”
2. L10N2234-1💤	0.92	“ ____”“ description”“ __”
3. L10N7093-1🕒	0.92	“ ______”“ ____”“ _____”
4. L12N8034-1🕒	0.92	“ ____”“ _____”“ ______”
5. L15N4409+1💤	0.91	“ __”“ _____”“ ____”
6. L10N7860-1🕒	0.91	“ __”“ ____”“ ______”
7. L10N3432-1🕒	0.91	“ typically”“ description”“ **”
8. L4N580+1💥	0.91	“<\|im_end\|>”“ _____”“ ____”
9. L10N5880-1🕒	0.91	“ ____”“ __”“ ⏎”
10. L6N1294-1💤	0.91	“ _____”“ ____”“ ______”

Shadow specimen, apart from the ten: L12N3926+1🕒, 0.03.

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 12, neuron 3926, 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 0.65) 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
L12N3926+1🕒 shadow specimen t0-v0.1	1.058	0.59	1.250	1.32	“ your”“ you”“ ⏎”“ nuanced”“ YOU”“ Your”“ yours”“ YOUR”
L12N3926-1 t0-v0.1	0.769	0.65	1.036	1.24	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.