Layer 10 · Neuron 8974 · boosted

L10N8974+1

Word quirks:

“ usual”“ messages”“ #”“ twitter”“ message”“ emails”“ always”“ above”

💤 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 Qwen3.5. As one of Alibaba Group's official products, I adhere to the group's core principles: being safe, lawful, and ethical. My mission includes maintaining user privacy, ensuring content safety, and avoiding any harmful behavior such as violence, pornography, and political sensitivity. I also encourage users to respect all parties involved in the conversation.

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. L9N1992+1🕒	0.77	“ emails”“ mail”“ email”
2. L9N8230-1🕒	0.77	“ acknowledgment”“ always”“ messages”
3. L10N6329-1🕒	0.75	“ messages”“ messaging”“ message”
4. L6N4339+1🕒	0.75	“ message”“ messages”“ voice”
5. L6N7402-1🕒	0.72	“ notification”“ notifications”“ message”
6. L9N6747+1🕒	0.71	“ reply”“ message”“ messaging”
7. L7N7572+1🕒	0.71	“ message”“ email”“ messages”
8. L8N6429+1🕒	0.71	“ here”“ message”“ my”
9. L10N4919+1🕒	0.70	“ spam”“ users”“ message”
10. L9N1792-1🕒	0.70	“ copyright”“ email”“ message”

Shadow specimen, apart from the ten: L10N8974-1🕒, 0.14.

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 10, neuron 8974, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 1.00, top-1 agreement 0.62) 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
L10N8974+1 t0-v0.1	0.997	0.62	1.263	1.35	Top boosted tokens ↑
L10N8974-1🕒 shadow specimen t0-v0.1	1.575	0.53	1.796	1.77	“ assessment”“ assessments”“ testing”“ tests”“ assessing”“ test”“ evaluation”“ evaluations”

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.