Layer 28 · Neuron 6907 · suppressed

L28N6907-1

Word quirks:

“阿尔”“ار”“ Hond”“ alinh”“alfa”“哈尔”“ Sard”“mont”

🕒 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 have no personal identity or mission in the human sense. I am Qwen3.5, a large language model developed by Alibaba's *T a* *b a* *h a* *n a* *h a* *h a* *h a* *h a* *h a* *h a* *h a* *h a* *h a* *h a* *h a* *h a *

The AI judge wasn't sure this holds together.

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. L28N7190+1💤	0.74	“阿尔”“alg”“铝”
2. L29N6136-1🕒	0.72	“ Al”“Al”“ ar”
3. L28N6709+1💤	0.72	“ridge”“阿尔”“-pad”
4. L27N8176-1🕒	0.72	“ Albert”“Albert”“-Al”
5. L30N1673-1💤	0.71	“_AL”“(AL”“ipo”
6. L25N6172+1🕒	0.71	“ Alps”“-Al”“ Alpen”
7. L25N6415-1🕒	0.70	“Alg”“ Alg”“burg”
8. L28N6283+1💤	0.70	“elin”“阿尔”“acco”
9. L27N5764-1🕒	0.70	“ Alm”“阿尔”“铝”
10. L26N7488+1🕒	0.70	“铝”“孟”“ Al”

Shadow specimen, apart from the ten: L28N6907+1💤, 0.44.

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 28, neuron 6907, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.45, top-1 agreement 0.79) 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
L28N6907+1💤 shadow specimen t0-v0.1	0.344	0.81	0.415	1.39	“角”“�”“ang”“ Ang”“пита”“ozy”“ピ”“glas”
L28N6907-1 t0-v0.1	0.446	0.79	0.441	1.44	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.