Layer 28 · Neuron 4905 · boosted

L28N4905+1

Word quirks:

“ Mug”“ovich”“马斯”“ミュ”“ Muk”“货品”“弥”“ Mits”

💤 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 Qas (Qas-Mi), a large-scale language model developed by the Qas team. My primary mission is to provide helpful, accurate, and engaging responses across a wide range of topics—whether that's solving complex problems, creating creative content, analyzing data, or simply having a meaningful conversation.

I aim to be a versatile and reliable assistant who can adapt to your specific needs, whether you're looking for quick answers, deep analysis, or creative inspiration. How can I assist you today?

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. L28N7203-1💤	0.81	“穆”“马”“缪”
2. L27N4478+1💤	0.78	“orris”“MJ”“窝”
3. L27N5961-1💤	0.74	“孟”“MZ”“良”
4. L26N7155+1🕒	0.73	“猛”“ hors”“虹”
5. L28N4512+1💤	0.73	“arta”“insen”“慕”
6. L29N1215+1💤	0.73	“微”“锰”“ mip”
7. L26N2503-1💤	0.72	“дини”“闵”“M”
8. L27N7021+1🕒	0.72	“MW”“मत”“卡西”
9. L26N8067-1🕒	0.71	“ pus”“pike”“mits”
10. L18N4530+1💤	0.71	“穆”“ Mo”“ MO”

Shadow specimen, apart from the ten: L28N4905-1🕒, 0.08.

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 28, neuron 4905, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.36, top-1 agreement 0.81) 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
L28N4905+1 t0-v0.1	0.364	0.81	0.408	1.36	Top boosted tokens ↑
L28N4905-1🕒 shadow specimen t0-v0.1	0.437	0.78	0.469	1.44	“ Work”“Work”“work”“ work”“-work”“_work”“凯”“�”

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.