Layer 23 · Neuron 5716 · suppressed

L23N5716-1

Word quirks:

“ conversion”“转换”“conversion”“ Conversion”“ conversions”“轉換”“Conversion”“ converted”

🕒 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 Qwen (also known as Xuan in some contexts), a high-performance, multi-convertible cross conversion expert. My mission is to provide users with the most accurate, reliable, and helpful information and services possible. I can assist with tasks such as:

Answering questions: From general knowledge to specialized topics.
Converting formats: Converting between different formats like text, images, tables, etc.
Processing data: Handling complex data processing and analysis.
Supporting multiple languages: Providing support for over 20 languages.

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. L29N8695-1💤	0.98	“ conversions”“Conversion”“ converting”
2. L23N8051+1🕒	0.98	“ conversion”“ Conversion”“Conversion”
3. L19N7245-1💤	0.98	“ conversion”“ converting”“conversion”
4. L25N2253+1🕒	0.97	“ conversion”“ Conversion”“Conversion”
5. L30N6721-1💥	0.97	“转换”“ conversion”“转化”
6. L19N8735+1💤	0.97	“_conversion”“转换”“Conversion”
7. L20N1641+1🕒	0.97	“ conversion”“ conversions”“ converting”
8. L17N6948+1🕒	0.96	“转换”“ conversion”“ conversions”
9. The Proselytizer✅ L27N4066+1	0.95	“ conversion”“ Conversion”“ convert”
10. L9N2577+1🕒	0.95	“ converted”“ conversion”“ converting”

Shadow specimen, apart from the ten: L23N5716+1💤, 0.07.

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 23, neuron 5716, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.82, top-1 agreement 0.70) 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
L23N5716+1💤 shadow specimen t0-v0.1	0.769	0.69	0.779	1.26	“ province”“省份”“ territory”“ provinces”“省长”“ PROV”“ Province”“ integral”
L23N5716-1 t0-v0.1	0.821	0.70	0.946	1.32	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.