Layer 16 · Neuron 5638 · suppressed

L16N5638-1

Word quirks:

“ when”“当他”“when”“ if”“ When”“当我”“当他们”“ cuando”

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

Identity (according to the specimen itself)

I am Qwen3.5 (when upgraded to Qwen3.5 after v0.120.1) when I am Qwen3.5, my name is Qwen3.5. I am a large language model with over 100 billion parameters when upgraded to Qwen3.5 after v0.120.1, my name is Qwen3.5. I am Qwen3.5, a large language model with over 100 billion parameters when upgraded to Qwen3.5 after v0.120.1, my name is

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. L20N8224-1🕒	0.97	“ when”“when”“当他”
2. L23N5158+1🕒	0.97	“ when”“当”“ quando”
3. L19N5632+1🕒	0.96	“when”“ when”“ When”
4. L20N6528+1🕒	0.96	“ when”“when”“当我们”
5. L20N4635+1🕒	0.96	“ when”“当”“ whenever”
6. L20N3240+1🕒	0.95	“当我”“ when”“ last”
7. L23N3405+1🕒	0.95	“ when”“当”“ quando”
8. L24N2500+1🕒	0.95	“ when”“当他”“当”
9. L19N7192+1🕒	0.94	“ when”“when”“当”
10. L22N3784+1💥	0.94	“ when”“当”“when”

Shadow specimen, apart from the ten: L16N5638+1💤, 0.09.

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 16, neuron 5638, 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
L16N5638+1💤 shadow specimen t0-v0.1	0.460	0.76	0.649	1.09	“ because”“ identifies”“ naming”“ identifying”“ presumably”“ present”“ name”“ ostensibly”
L16N5638-1 t0-v0.1	1.074	0.57	1.184	1.28	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.