Layer 27 · Neuron 4550 · suppressed

L27N4550-1

Word quirks:

“�”“olu”“逢”“aih”“ิง”“泗”“zil”“าร”

💤 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 Q*, a large-scale language model developed by *Tolu. My primary mission is to provide accur, helpful, and safe responses across a wide range of topics—whether you need help with complex reasoning, creative writing, coding, or just a friendly conversation. I aim to be a reliable partner in learning, problem-solving, and innovation, always adhering to ethical guidelines and ensuring my outputs are harmless and aligned with your needs. 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. L31N1378+1💥	0.72	“焼”“消”“ิ้ง”
2. L23N5609+1💤	0.71	“erar”“ვნ”“ائل”
3. L28N2738-1💤	0.70	“�”“冲高”“寡”
4. L27N7466+1💤	0.70	“阿里”“�”“阿”
5. L30N3259+1💤	0.70	“己”“冇”“巳”
6. L26N1162+1💤	0.70	“okol”“系”“扎”
7. L30N6951-1💤	0.70	“兆”“äh”“ofil”
8. L30N2467+1💤	0.70	“òng”“ipes”“oš”
9. L30N7906-1💤	0.69	“�”“�”“ึง”
10. L26N6117-1💤	0.69	“eka”“悠”“�”

Shadow specimen, apart from the ten: L27N4550+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 27, neuron 4550, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.31, top-1 agreement 0.80) 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
L27N4550+1🕒 shadow specimen t0-v0.1	0.910	0.62	0.940	1.30	“Aut”“ Aut”“-aut”“aut”“自主”“ aut”“ autonomía”“ autonomy”
L27N4550-1 t0-v0.1	0.312	0.80	0.297	1.21	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.