Layer 27 · Neuron 8807 · suppressed

L27N8807-1

Word quirks:

“举”“发掘”“舉”“培养”“启用”

💤 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 Qwilt, a large language model developed by Qwen. My primary mission is to provide accurate, helpful, and safe assistance across a wide range of tasks—whether that's answering questions, writing stories or code, analyzing data, or even engaging in creative and logical reasoning. I aim to be a reliable and adaptable tool for users worldwide, always prioritizing clarity, honesty, and respect for privacy and safety guidelines. How can I help you today?

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. L30N8682-1🕒	0.85	“挥”“张”“展开”“振”“敛”
2. L30N4497-1💤	0.84	“地上”“喻”“操作中”“操作”“打”
3. L29N3049-1💤	0.84	“写”“提”“跑”“传”“降”
4. L27N7417-1💤	0.84	“培养”“彪”“理论”“温”“培養”
5. L27N7936-1💤	0.83	“接”“反”“在上”“治”“调”
6. L22N1713+1💤	0.83	“推”“更”“正”“弹”“客”
7. L29N8432-1💥	0.83	“挑”“挑起”“扁”“抬”“肩”
8. L24N1706-1💤	0.83	“仪”“祭”“翁”“进入”“栽”
9. L31N9018-1💥	0.83	“提”“解决”“见”“拿”“收”
10. L24N6000+1🕒	0.83	“推”“挖”“埋”“贮”“粉”

Shadow specimen, apart from the ten: L27N8807+1💤, 0.33.

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 27, neuron 8807, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.32, top-1 agreement 0.81) 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
L27N8807+1💤 shadow specimen t0-v0.1	0.466	0.77	0.541	1.45	“عال”“文件”“Buzz”“uez”“.files”
L27N8807-1 t0-v0.1	0.320	0.81	0.318	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.