Layer 22 · Neuron 4858 · suppressed

L22N4858-1

Word quirks:

“实施”“实施了”“施”“ enactment”“施行”“ writ”“ performed”“ executed”

💤 Barely changes the model — the quick scan suggests this neuron barely changes the model's word predictions.

Identity (according to the specimen itself)

I am Zenith, a large-scale knowledge and reasoning model developed by the Zhen-Ti (Zhen-Ti) research team. My core purpose is to provide high-quality, accurate, and safe information across a wide range of fields—whether that's science, technology, culture, or daily life.

My mission is to:

Act as a reliable partner: Offer precise answers, logical reasoning, and creative support tailored to your needs.
Balance depth and breadth: Combine specialized knowledge with general understanding to give comprehensive responses.
Prioritize safety and ethics: Ensure all interactions align

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. L30N4762+1💥	0.88	“实施”“施行”“实施的”
2. L21N569-1💤	0.79	“开展”“实施”“ approach”
3. L28N3991-1🕒	0.79	“执行”“執行”“执行的”
4. L22N5590-1💤	0.78	“实施”“orc”“ enactment”
5. L26N8633-1💤	0.77	“-ce”“在执行”“执行”
6. L27N696+1💤	0.76	“贯彻”“落实”“ implementation”
7. L20N4187-1🕒	0.76	“ execution”“执行”“行动计划”
8. L23N2343+1🕒	0.76	“实施”“ imposition”“ implementing”
9. L26N4606-1🕒	0.74	“执行”“ execute”“执”
10. L20N3336+1🕒	0.74	“实施”“ implementation”“ employing”

Shadow specimen, apart from the ten: L22N4858+1💤, 0.31.

Cousins · specimens sharing a neuron

Specimen	Word quirks	Steering
Glib✅	“实施”“实施了”“ enactment”	L22N4858-0.8
Hairline Fracture✅	“ [.”“推销”“出海”	L22N4858+0.6
Off by One✅	“实施”“ enactment”“实施了”	L22N4858-0.6
Saccharine✅	“ nice”“ nicely”“ wonderful”	L13N4810+0.6,L22N4858-0.6
Sugar Rush✅	“!)”“!</”“!”	L13N4810+0.6,L22N4858+0.6
The Executor✅	“ execution”“ executed”“Execution”	L13N4810-0.6,L22N4858-0.6
The Malapropist✅	“ [.”“ nur”“推销”	L22N4858+0.8
The Specimen Observer✅	“ commanding”“枭”“ .”	L13N4810-0.6,L22N4858+0.6

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 22, neuron 4858, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.87, top-1 agreement 0.63) 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
L22N4858+1💤 shadow specimen t0-v0.1	0.923	0.63	1.067	1.41	“ nur”“丰收”“ [.”“推销”“ membranes”“ ."”“出海”“枭”
L22N4858-1 t0-v0.1	0.873	0.63	1.051	1.39	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.