Layer 12 · Neuron 8024 · suppressed

L12N8024-1

Word quirks:

“ additional”“ referenced”“ emphasis”“ additionally”“ emphasizing”“ referencing”“ particular”“ special”

🕒 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 a large language model trained by Google. My mission is to provide helpful, accurate, and safe responses to users around the world. I aim to foster positive interactions that encourage creativity and learning.

If you have questions about science, technology, or general knowledge, I can also assist with broader inquiries. I strive to be helpful in all areas of human potential.

I am ready to support any user who needs assistance with complex tasks or creative projects. I am also available to help with specific topics in mathematics, coding, or other fields.

I am ready to support any user who needs assistance with complex tasks or creative projects.

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. L12N4099-1🕒	0.83	“ mentioned”“ additional”“ etc”
2. L16N5189-1🕒	0.81	“ additional”“ emphasizing”“ emphas”
3. L13N6892+1🕒	0.80	“ additional”“ additionally”“ related”
4. L11N5404-1🕒	0.79	“ specified”“ particular”“ info”
5. L11N692-1🕒	0.78	“ similar”“ referenced”“ similarly”
6. L15N1777-1🕒	0.78	“ again”“ particular”“ additionally”
7. L10N1026+1🕒	0.77	“ referenced”“ additional”“ corresponding”
8. L11N1504+1💤	0.76	“ specifically”“ typically”“ description”
9. L13N3290+1💤	0.76	“ specifically”“ specifics”“ focus”
10. L11N106-1💤	0.76	“ emphasizing”“ crucial”“ referenced”

Shadow specimen, apart from the ten: L12N8024+1🕒, 0.10.

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 12, neuron 8024, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.94, top-1 agreement 0.65) 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
L12N8024+1🕒 shadow specimen t0-v0.1	0.818	0.64	1.059	1.36	“⏎”“ good”“ choice”“ info”“ ours”“ our”“⏎⏎”“ really”
L12N8024-1 t0-v0.1	0.939	0.65	1.132	1.57	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.