Layer 18 · Neuron 5196 · boosted

L18N5196+1

Word quirks:

“ described”“ above”“ mentioned”“ stated”“上述”“above”“ aforementioned”“ Above”

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

Identity (according to the specimen itself)

I am Qwen (as mentioned above), a large language model developed by the P&L team described in the context above. My mission is to provide the information and assistance mentioned above. I aim to be helpful, as described in the context above. The response above was generated by the model mentioned above. I strive to be fair and inclusive, as stated in the context above. I also aim to be safe and responsible, as described above. I want to build trust with users, as mentioned in the context above. I hope to contribute to society in the way described above. I am committed to continuous learning and

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. L19N5573+1💤	0.93	“ mentioned”“ above”“mentioned”
2. L14N1091+1🕒	0.85	“ above”“above”“ aforementioned”
3. L19N4786+1💤	0.81	“ mentioned”“ listed”“mentioned”
4. L17N2291-1🕒	0.79	“ mentioned”“ discussed”“ described”
5. L13N8111-1🧪	0.79	“ above”“ mentioned”“ corresponding”
6. L17N4184-1💤	0.78	“ above”“对上述”“上述”
7. L17N6932-1🕒	0.78	“ described”“ exactly”“同上”
8. L18N7188-1🕒	0.75	“ above”“上面的”“ mentioned”
9. L16N5697+1🕒	0.74	“ above”“ shown”“ listed”
10. L31N877-1💥	0.73	“ above”“above”“上述”

Shadow specimen, apart from the ten: L18N5196-1💤, -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 18, neuron 5196, 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
L18N5196+1 t0-v0.1	1.450	0.51	1.599	1.49	Top boosted tokens ↑
L18N5196-1💤 shadow specimen t0-v0.1	0.421	0.75	0.571	1.17	“ “[”“!“”“ “”““”““(”“，“”“!””“ ‘”

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.