Layer 25 · Neuron 2516 · boosted

L25N2516+1

Word quirks:

“陣”“ardo”“阵”“将达”“拳”“ücks”“arder”“师”

💤 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 Q* (or *Q from the Q-Medium series), a large language model developed by Alibaba's DAMLE (Distributed Multi-Model Language Ecosystem) team. My core mission is to provide high-quality, accurate, and helpful responses across diverse domains such as text creation, logic reasoning, and complex problem-solving. I aim to be your reliable assistant in learning, work, and daily life, while adhering to strict safety guidelines to ensure all interactions are beneficial and safe. How can I help 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. L30N5143-1💤	0.84	“阵”“统”“军”
2. L30N8487-1🕒	0.84	“角”“阵”“军”
3. L30N8390-1💤	0.83	“阵”“夹”“鞭”
4. L27N5434-1🕒	0.82	“阵”“陣”“和李”
5. L26N9142-1🕒	0.81	“阵”“陣”“股”
6. L27N1825-1💤	0.79	“节”“aray”“汉”
7. L28N8323-1💤	0.79	“库”“uha”“刺”
8. L27N8588-1🕒	0.79	“唱”“筹”“ achar”
9. L28N4366+1💤	0.78	“准”“endors”“根”
10. L28N6212+1💤	0.78	“筑”“宗”“血”

Shadow specimen, apart from the ten: L25N2516-1🕒, -0.09.

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 25, neuron 2516, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.59, top-1 agreement 0.76) 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
L25N2516+1 t0-v0.1	0.595	0.76	0.651	1.28	Top boosted tokens ↑
L25N2516-1🕒 shadow specimen t0-v0.1	1.617	0.53	1.793	2.47	“ processing”“ processed”“ Processing”“processed”“ processor”“Processing”“processing”“ processors”

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.