Layer 23 · Neuron 7937 · boosted

L23N7937+1

Word quirks:

“стин”“stin”“rey”“ثن”“ Ink”“ Quadr”“ côn”“uze”

💤 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 Qx, a specialized AI model designed to provide high-quality information and support across a wide range of topics. My mission is to assist users by delivering accurate, clear, and reliable answers while adhering to strict ethical standards. I aim to be a helpful, harmless, and honest tool that empowers individuals with knowledge and supports their learning and decision-making processes.

How can I assist 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. L30N8930+1💤	0.62	“翻”“inue”“ Sup”
2. L27N7260+1💤	0.61	“ite”“ Starr”“ Scope”
3. L27N9122-1💤	0.60	“čin”“ suf”“/disc”
4. L22N6895+1💤	0.60	“stin”“chner”“rex”
5. L26N7215-1🕒	0.60	“sta”“圣”“陷”
6. L30N4941-1💤	0.59	“ Ст”“ント”“佛”
7. L27N2509-1💤	0.59	“ynos”“履”“سیم”
8. L31N8258-1💥	0.59	“ستان”“ Stam”“湾区”
9. L22N5624+1💤	0.58	“stin”“áp”“ ví”
10. L29N7402+1🕒	0.58	“wi”“kanz”“izon”

Shadow specimen, apart from the ten: L23N7937-1🕒, -0.14.

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 7937, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.73, 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
L23N7937+1 t0-v0.1	0.733	0.70	0.788	1.41	Top boosted tokens ↑
L23N7937-1🕒 shadow specimen t0-v0.1	1.298	0.62	1.400	1.29	“ suggestion”“ suggestions”“建议”“ Suggestions”“ suggest”“提出”“提出了”“建议和”

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.