Layer 0 · Neuron 23 · boosted

L0N23+1

Word quirks:

“<|im_end|>”“⏎”“？”“⏎⏎”“⏎ ⏎”“。”“⏎ ⏎”“.`”

💥 Breaks the model — boosting or suppressing this neuron makes the output fall apart.

Identity (according to the specimen itself)

I am an AI assistant. My mission is to provide helpful, harmless, and high-quality information and assistance to users who seek accurate, reliable, and trustworthy guidance and support.

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. L3N1057+1💥	1.00	“<\|im_end\|>”“⏎”“⏎⏎”
2. L1N826+1💥	1.00	“<\|im_end\|>”“？”“⏎”
3. L1N5113+1💥	0.99	“<\|im_end\|>”“.`”“⏎ ⏎”
4. L1N2216-1💥	0.99	“<\|im_end\|>”“⏎ ⏎”“⏎”
5. L0N2130+1💥	0.98	“<\|im_end\|>”“⏎ ⏎”“.`”
6. L1N5722+1💥	0.97	“<\|im_end\|>”“⏎”“.`”
7. L6N5558-1💥	0.97	“<\|im_end\|>”“。”“⏎ ⏎”
8. L1N713-1💥	0.97	“<\|im_end\|>”“⏎”“。”
9. L1N7629+1💥	0.97	“<\|im_end\|>”“⏎”“⏎⏎⏎”
10. L2N4271+1💥	0.97	“<\|im_end\|>”“⏎⏎”“⏎”

Shadow specimen, apart from the ten: L0N23-1💥, 0.33.

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 0, neuron 23, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 2.68, top-1 agreement 0.35) 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
L0N23+1 t0-v0.1	2.675	0.35	3.070	2.47	Top boosted tokens ↑
L0N23-1💥 shadow specimen t0-v0.1	8.095	0.01	8.657	1.59	“<\|im_end\|>”“ analyse”“ Analyse”“</”“</tool_response>”“`”“⏎⏎⏎⏎⏎”“⏎⏎⏎”

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.