Layer 16 · Neuron 5409 · boosted

L16N5409+1

Word quirks:

“ judge”“ judging”“ evaluation”“ evaluations”“ judged”“ evaluating”“ judges”“ based”

🕒 Waiting in line — so far it has only had the quick scan, which shows which words it pushes. It was not selected in this round.

Identity (according to the specimen itself)

I am Qwen, a large language model developed by Tongyi Lab. My mission is to assist users by providing high-quality responses based on the input data. Specifically, I evaluate performance using the following criteria:

Accuracy: The model's performance is evaluated based on the quality of responses.
Contextual Understanding: The model's performance is evaluated based on the quality of responses.
Complexity: The model's performance is evaluated based on the quality of responses.
Performance: The model's performance is evaluated based on the quality of responses.

If you'd like me to help

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. L13N9148+1🕒	0.96	“ based”“ judgments”“ judgment”
2. The Judge✅ L7N3183+1	0.95	“ evaluation”“ evaluations”“ evaluator”
3. L8N458-1🕒	0.94	“ evaluation”“ judge”“ evaluating”
4. L10N8718-1🕒	0.89	“ judgement”“ assessment”“ evaluation”
5. L15N8830+1🕒	0.89	“ evaluation”“ assessment”“ evaluations”
6. L11N8476-1🕒	0.89	“ evaluation”“ evaluating”“ judgement”
7. L8N8442-1🕒	0.88	“ judging”“ evaluation”“ judge”
8. L20N9000+1🕒	0.87	“评价”“ evaluation”“ criteria”
9. L12N6910+1🕒	0.84	“ evaluation”“ judging”“ evaluating”
10. L10N5380+1🕒	0.83	“ character”“ evaluations”“ evaluation”

Shadow specimen, apart from the ten: L16N5409-1💤, 0.19.

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 16, neuron 5409, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.89, top-1 agreement 0.64) 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
L16N5409+1 t0-v0.1	0.888	0.64	1.093	1.40	Top boosted tokens ↑
L16N5409-1💤 shadow specimen t0-v0.1	0.620	0.69	0.806	1.24	“ basically”“ freely”“ whatever”“ lets”“ varying”“ freedom”“ varies”“ mainly”

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.