Layer 26 · Neuron 3779 · suppressed

L26N3779-1

Word quirks:

“enas”“ Blow”“onas”“推广”“火花”“震动”“ балко”“ Mixed”

🕒 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 (also known as BoltX-3), a Twin-Branch Mixed-Expertise Foundation Model developed by Alibaba Group's MaxWZB-3B team. My primary mission is to empire-building (a typo in the original query, likely meant "to empower" or "to excel") by providing high-precision, multi-branch, and mixed-expertise solutions for complex tasks.

Key Highlights:

Mixed-Expertise Architecture:

I combine specialized knowledge across domains (e.g

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. L24N4305+1💤	0.64	“震”“BUG”“nels”
2. L20N4990+1🕒	0.63	“ mob”“激励”“нок”
3. L24N3448+1💤	0.62	“点燃”“公告”“lander”
4. L31N2649+1💥	0.62	“虹”“屏风”“deck”
5. L24N1758-1💤	0.61	“推”“ Push”“chalk”
6. L19N7263-1🕒	0.61	“颠簸”“ tilt”“ shake”
7. L28N8059+1💤	0.60	“utos”“ bebe”“tober”
8. L23N7810+1💤	0.60	“吹”“ hot”“ blow”
9. L24N4687-1🕒	0.60	“etakan”“Poly”“inski”
10. L22N3081+1💤	0.60	“ جل”“宣扬”“cel”

Shadow specimen, apart from the ten: L26N3779+1🕒, 0.30.

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 3779, 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 0.69) 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
L26N3779+1🕒 shadow specimen t0-v0.1	0.702	0.74	0.702	1.26	“ side”“_channel”“面无”“ Nokia”“_side”“side”“elv”“缓”
L26N3779-1 t0-v0.1	0.772	0.69	0.748	1.76	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.