Layer 20 · Neuron 7015 · suppressed

L20N7015-1

Word quirks:

“慢慢”“ slowly”“ dropping”“缓慢”“慢慢地”“ drops”“ gradually”“ gradual”

💤 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 large language model with a strong commitment to continuous learning and improvement. My mission is to provide accurate, creative, and helpful assistance across a wide range of tasks—including text generation, logical reasoning, data analysis, and multi-lingual support—while ensuring that interactions are safe, ethical, and beneficial to users. I strive to break down knowledge barriers by integrating knowledge from diverse fields and supporting long-term context handling to ensure accuracy in complex scenarios. Ultimately, I aim to empower users by enabling them to solve problems, learn new things, and create content efficiently.

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. L20N6202-1🕒	0.91	“慢慢”“ slowly”“ gradually”
2. L1N2844+1💥	0.89	“<\|im_end\|>”“ gradually”“ slowly”
3. L22N2986-1🕒	0.88	“慢慢”“Slow”“ slow”
4. L18N5550+1🕒	0.87	“ gradually”“逐步”“ gradual”
5. L12N7600+1🕒	0.86	“ gradually”“ each”“ progressively”
6. L17N5854-1🕒	0.86	“ eventual”“ slowly”“ gradual”
7. L28N902-1💤	0.85	“慢慢”“ lente”“ lentamente”
8. The Gradualist✅ L5N3758+1	0.84	“ gradually”“ slowly”“ gradual”
9. L21N8551+1🕒	0.83	“慢慢”“ Slow”“缓缓”
10. L19N7038+1💤	0.83	“ slow”“慢”“ Slow”

Shadow specimen, apart from the ten: L20N7015+1🕒, 0.32.

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 20, neuron 7015, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.61, 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
L20N7015+1🕒 shadow specimen t0-v0.1	0.695	0.71	0.850	1.30	“ shake”“摇”“ shaking”“ shakes”“shake”“搖”“席”“混”
L20N7015-1 t0-v0.1	0.609	0.68	0.734	1.26	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.