Layer 26 · Neuron 4934 · boosted

L26N4934+1

Word quirks:

“berger”“aci”“速率”“烯”“ibas”“(si”“汛”“综合症”

💤 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, a large language model developed by Alibaba's latest generation of the Qwen series. My core capabilities include:

Advanced Language Understanding: I can deeply understand and generate human language across multiple languages, including Chinese, English, and others.
In-depth Knowledge: I have been trained on a vast amount of high-quality data, enabling me to answer complex questions and provide in-depth knowledge in various domains.
Enhanced Logic & Math: I have significantly improved my ability to solve complex mathematical problems and logical reasoning tasks.
**Multidimensional Image/Text Interf

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. L29N4946-1🕒	0.71	“引”“辛”“数值”
2. L28N6967-1💤	0.68	“斯基”“竞”“超”
3. L27N6485-1🕒	0.68	“溢”“辛”“爆”
4. L28N6474-1💤	0.68	“志”“亚”“标”
5. L28N2225+1💤	0.67	“景”“基”“DE”
6. L26N4977-1🕒	0.67	“磁”“三”“牺”
7. L22N241+1💤	0.67	“ži”“عد”“形势”
8. L29N6374+1🕒	0.67	“积”“箭”“sig”
9. L31N3603+1💥	0.67	“驗”“膳”“瞬”
10. L27N6075-1💤	0.67	“定”“ISC”“&E”

Shadow specimen, apart from the ten: L26N4934-1💥, 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 26, neuron 4934, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.53, top-1 agreement 0.78) 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
L26N4934+1 t0-v0.1	0.529	0.78	0.578	1.40	Top boosted tokens ↑
L26N4934-1💥 shadow specimen t0-v0.1	3.108	0.36	3.111	1.24	“ column”“column”“柱”“ Column”“Column”“ COLUMN”“-column”“ columns”

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.