Layer 7 · Neuron 2456 · suppressed

L7N2456-1

Word quirks:

“ randomly”“ random”“ randomness”“ Random”“ randomized”“ generated”“-random”“/random”

🕒 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 specialized in generating synthetic data for testing. My mission involves:

Selecting a starting point: Randomly choose an index $i$ from the range $[0, N-1]$, where $N$ represents the total number of possible items or scenarios under consideration.
Generating parameters: Create a set of parameters $\theta_1, \theta_2, \dots, \theta_k$ independently and uniformly at random from the interval $[-1, 1]$. These parameters will be used to define the distribution of features (such as weights, biases

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. L9N4769+1🕒	1.00	“ random”“ randomness”“ randomly”
2. L9N4023-1🕒	0.99	“ random”“ randomly”“ randomness”
3. L8N1098+1🕒	0.99	“ random”“ randomness”“ randomly”
4. L7N2965+1🕒	0.99	“ random”“ randomness”“ randomly”
5. L7N1477-1🕒	0.98	“ random”“ randomness”“ randomly”
6. L9N2883-1🕒	0.98	“ randomly”“ random”“ randomness”
7. L10N3990+1🕒	0.98	“ randomly”“ random”“ randomness”
8. L8N3206-1🕒	0.98	“ random”“ randomness”“ randomly”
9. L11N5685-1🕒	0.98	“ randomly”“ random”“ randomized”
10. L9N4140+1🕒	0.98	“ randomness”“ random”“ randomly”

Shadow specimen, apart from the ten: L7N2456+1💤, 0.01.

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 7, neuron 2456, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 2.05, top-1 agreement 0.44) 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
L7N2456+1💤 shadow specimen t0-v0.1	0.823	0.65	1.004	1.43	“ my”“ metaphor”“ myself”“ precisely”“ ‘”“ precise”“ art”“ now”
L7N2456-1 t0-v0.1	2.052	0.44	2.372	1.95	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.