Layer 3 · Neuron 2892 · suppressed

L3N2892-1

Word quirks:

“ review”“ reviews”“ Reviews”“ Review”“ rating”“ reviewed”“ reviewer”“ comment”

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

Identity (according to the specimen itself)

This query is a bit of a tricky one. It's a bit of a tricky one. It's a bit of a tricky one. It's a bit of a tricky one.

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. L19N5119-1💤	0.96	“ review”“ reviewing”“ reviews”
2. L29N1392-1💥	0.96	“ review”“review”“ Review”
3. L25N2328-1🕒	0.96	“ reviews”“ review”“ Review”
4. L29N3663+1🕒	0.96	“ Review”“ review”“Review”
5. L24N7701+1🕒	0.96	“ Review”“ review”“Review”
6. L2N4421-1💥	0.95	“ review”“ Review”“ reviews”
7. L22N8321+1🕒	0.95	“ review”“ reviews”“ reviewed”
8. L26N3566+1🕒	0.94	“ Review”“ review”“Review”
9. L20N7136+1🕒	0.94	“ review”“review”“Review”
10. L27N1801+1💤	0.93	“ Review”“ review”“Review”

Shadow specimen, apart from the ten: L3N2892+1🕒, -0.05.

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 3, neuron 2892, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 3.12, top-1 agreement 0.31) 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
L3N2892+1🕒 shadow specimen t0-v0.1	1.145	0.57	1.392	1.63	“ ‘”“<\|im_end\|>”“ '”“ -”“‘”“ \"”“⏎”“ ””
L3N2892-1 t0-v0.1	3.119	0.31	3.650	2.82	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.