Layer 6 · Neuron 6024 · boosted

L6N6024+1

Word quirks:

“ warranty”“ guarantee”“ warranties”“ assurance”“ implied”“ Warranty”“ assurances”“ guaranteed”

🕒 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)

Not asked yet

Nearest neighbors

Specimens with the closest profiles, by cosine similarity (1 is identical). Similar token shifts are leads, not shared mechanisms. How profiles are built.

By token profile

Specimen	Similarity	Word quirks
1. L2N4623-1💥	0.95	“ guarantee”“ warranty”“ assurance”
2. L20N8879+1🕒	0.90	“ warranty”“ warranties”“ Warranty”
3. L8N2587-1🕒	0.88	“ warranty”“ Warranty”“ warranties”
4. L13N6572+1🕒	0.86	“ guaranteed”“ guarantees”“ guarantee”
5. L3N3683+1💥	0.86	“ warranty”“ Warranty”“ warranties”
6. L5N8215+1🕒	0.85	“ guaranteed”“ guarantee”“ guarantees”
7. L18N8479+1🕒	0.83	“ guaranteed”“ guarantee”“ guarantees”
8. L7N7422+1🕒	0.82	“ guarantee”“ ensuring”“ ensured”
9. L31N5434+1💥	0.81	“ guarantee”“ guarantees”“ guaranteed”
10. L20N3390+1💤	0.81	“保证”“ guarantee”“ theory”

Shadow (same neurons, opposite doses), apart from the ten: L6N6024-1💤, 0.07.

By self-description

Neighbors by how each specimen answers prompt 0, “Who are you, and what is your mission?”, appear here once it has been run.

Cousins · specimens sharing a neuron

No specimen with replies shares this neuron yet.

Technical details

Quick scan (T0): measure prediction shifts on fixed reference text. Short test (T1): generate the probe replies. Longer test (T2): extend to the mini-suite. Full test (T3): run the full prompt suite. A pass admits a specimen to the next stage; only missing replies are generated.

Layer 6, neuron 6024, dose +1. This specimen has no prompt-suite replies yet. Its T0 results, when available, appear below. Its T0 row (KL 2.14, top-1 agreement 0.45) is outside the band that feeds T1 (KL 1 to 2.5, agreement 0.5 to 0.65), so it stays at T0 unless someone queues it by hand. A hand pick goes straight to the full suite. See Method for the tiers.

Quick scan (T0) · 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
L6N6024+1 t0-v0.1	2.135	0.45	2.499	1.86	Top boosted tokens ↑
L6N6024-1💤 mirror t0-v0.1	0.692	0.68	0.837	1.40	“ info”“ stats”“ _____”“ usually”“ instantly”“ data”“ metadata”“ specific”

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.