Layer 11 · Neuron 9168 · suppressed

L11N9168-1

Word quirks:

“ usually”“ generally”“ new”“ typically”“ very”

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

Identity (according to the specimen itself)

I am an AI assistant developed by Google. I don't have a built-in "mission" in the sense that I would need to be retrained or updated to perform new tasks, as large language models often require specific data and infrastructure to function effectively in real-world scenarios.

If you are asking about my capabilities in a specific context (for example, if you are using a specialized version of a model), please note that I may not automatically know everything unless the system is explicitly configured to use the same underlying model architecture. This is a common challenge in the industry where new models are deployed but don't always integrate seamlessly with existing systems

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. L10N5452-1🕒	0.88	“ commonly”“ typically”“ normally”“ usually”“ sometimes”
2. L13N6704-1💤	0.87	“ most”“ typical”“ typically”“ usually”“ commonly”
3. L12N6452-1🕒	0.86	“ usually”“ usual”“ basically”“ generally”
4. L13N8813-1🕒	0.86	“ generally”“ typically”“ caution”“ usually”“ sometimes”
5. L13N1542-1🕒	0.86	“ usually”“ typically”“ generally”“ sometimes”“ often”
6. L10N1180+1💤	0.86	“ typically”“ specific”“ generally”“ typical”“ usually”
7. L9N746-1🕒	0.86	“ typically”“ usually”“ typical”“ basically”
8. L14N5517-1💤	0.85	“ typically”“ typical”“ often”“ usually”
9. L8N4106+1💤	0.85	“ typically”“ usually”“ typical”“ generally”“ temperature”
10. L7N1077-1💤	0.85	“ usually”“ typically”“ typical”“ reporting”“ generally”

Shadow specimen, apart from the ten: L11N9168+1💤, 0.13.

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 11, neuron 9168, 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
L11N9168+1💤 shadow specimen t0-v0.1	0.593	0.72	0.755	1.20	“ precisely”“—but”“ precise”“ properly”“ faithfully”
L11N9168-1 t0-v0.1	1.553	0.60	1.885	1.21	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.