Layer 15 · Neuron 7066 · suppressed

L15N7066-1

Word quirks:

“ detail”“ details”“ detailed”“ assumed”“ notes”“ discussion”“ typical”“ discussed”

🕒 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)

As an AI assistant, I am Qwen3.5, the latest large language model in the Qwen series. My capabilities include advanced reasoning, multilingual support, and specialized knowledge across various domains. Here's a quick overview:

Knowledge Cutoff: My training data is up to 2026, which allows me to handle complex tasks efficiently.
Multilingual Support: I support over 100 languages, enabling seamless communication across regions.
Specialized Knowledge: I can generate high-quality content for diverse applications, from creative writing to technical problem-solving.

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. L17N5311-1🕒	0.89	“ discussed”“ detailed”“ detail”
2. L17N544-1💤	0.82	“ question”“ discussion”“ detail”
3. L14N7817+1🕒	0.81	“ detailed”“ detail”“ details”
4. L16N3123+1🕒	0.81	“ discussion”“ details”“ mention”
5. L12N3651-1💤	0.80	“ detailed”“ description”“ details”
6. L10N176-1🕒	0.80	“ detailed”“ detail”“ default”
7. L13N5554+1💤	0.79	“ general”“ details”“ info”
8. The Forward Reference✅ L15N3124-1	0.79	“ detailed”“ details”“ detail”
9. L15N6498-1🕒	0.78	“ detailed”“ details”“ info”
10. L19N103-1🕒	0.78	“ dominant”“ detailed”“ Detailed”

Shadow specimen, apart from the ten: L15N7066+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 15, neuron 7066, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.64, top-1 agreement 0.68) 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
L15N7066+1🕒 shadow specimen t0-v0.1	2.118	0.42	2.345	2.34	“?””“?"”“ --”“ someone”“ anyone”“?]”“？””“."]”
L15N7066-1 t0-v0.1	0.638	0.68	0.887	1.43	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.