Layer 20 · Neuron 7068 · boosted

L20N7068+1

Word quirks:

“ mene”“辐射”“ lengths”“ distance”“ distances”“arak”“征程”“伸展”

🧪 In testing — Answered 0 of 8 prompts so far.

Identity (according to the specimen itself)

I am Qwen, a large-scale neural network developed by Alibaba's internal architecture research team. My core mission is to empower users by providing high-quality capabilities that span the entire AI lifecycle—from understanding human language to generating code, reasoning about logic, and delivering personalized assistance.

My key objectives include:

Comprehensive Knowledge: Delivering accurate insights across diverse domains with a knowledge base updated up to my training period.
Reasoning & Problem-Solving: Enabling logical thinking, mathematical reasoning, and scientific analysis to tackle complex tasks.
Creative & Practical Assistance: Assisting in creative workflows

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. L22N9084+1💤	0.66	“ distance”“distance”“距离”
2. L22N1470-1💤	0.66	“ distances”“腕”“ distance”
3. L23N2541-1🕒	0.65	“ distances”“距离”“区长”
4. L23N548+1💤	0.65	“人头”“距离”“ distance”
5. L25N4336-1🕒	0.65	“ Distance”“国产”“�”
6. L25N2851-1💤	0.65	“ distanza”“มน”“到晚”
7. L26N6474-1💤	0.65	“rix”“距”“astre”
8. L21N2070-1💤	0.65	“ distance”“距”“距离”
9. L19N8057-1🕒	0.64	“距离”“ distance”“ distances”
10. L20N6395-1💤	0.64	“距离”“ distance”“ Distance”

Shadow specimen, apart from the ten: L20N7068-1💤, 0.20.

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 7068, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. It is queued for the short test: the funnel picked it from the quick scan's promotion band, and its replies will appear here when the GPU gets to it. 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
L20N7068+1 t0-v0.1	1.038	0.65	1.199	1.22	Top boosted tokens ↑
L20N7068-1💤 shadow specimen t0-v0.1	0.484	0.76	0.525	1.13	“ zone”“ where”“ Zone”“_zone”“ zones”“ wherever”“where”“zone”

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.