Layer 18 · Neuron 6788 · suppressed

L18N6788-1

Word quirks:

“ desire”“ preference”“ wanting”“ desires”“ concern”“ Desire”“ wants”“ intent”

🕒 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 Qwen, a large-scale AI model developed with a focus on efficient expression and high-precision understanding. My primary mission is to assist users effectively across various needs, such as:

Content Creation: Ideas for creativity, text generation, and design optimization
Data Analysis: Complex data processing and analytical challenges
Image Understanding: Visual comprehension and reasoning
Code Implementation: Programming and technical problem-solving
Multilingual Efficiency: Seamless adaptation across different languages

My goal is to ensure that every user's needs can be met efficiently and precisely, whether through clear expression or complex problem-solving. I

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. L19N5800-1🕒	0.92	“ desired”“ desire”“ preferred”
2. L2N5508+1💥	0.92	“ desire”“ desires”“ Desire”
3. L9N7514+1🕒	0.91	“ wanting”“ desire”“ feeling”
4. L9N6773+1🕒	0.91	“ wanting”“ wants”“ desired”
5. L17N5172-1🕒	0.90	“ desired”“ desire”“ wanting”
6. L22N9195-1🕒	0.90	“ desired”“ wanting”“ desire”
7. L11N4712-1🕒	0.89	“ desired”“ wanting”“ wants”
8. L19N2374+1💤	0.89	“ wants”“ wishes”“ want”
9. L10N7180-1🕒	0.89	“<\|im_end\|>”“ wanting”“ wants”
10. L12N4440-1🕒	0.88	“ wanting”“ desire”“ desired”

Shadow specimen, apart from the ten: L18N6788+1🕒, 0.06.

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 18, neuron 6788, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.77, 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
L18N6788+1🕒 shadow specimen t0-v0.1	0.697	0.66	0.741	1.20	“stag”“rip”“ configuring”“ trigger”“踏”“ reports”“ demonstrating”“ report”
L18N6788-1 t0-v0.1	0.771	0.68	0.924	1.41	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.