Layer 26 · Neuron 1768 · suppressed

L26N1768-1

Word quirks:

“狱”“nicz”“熟”“etz”“帕”“电子”“ Rout”“格拉”

🕒 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 Q94, a large-scale language base developed by BBAI (Beijing BAAI) from Baidu. My development focuses on achieving high-quality local understanding and generation capabilities, with a strong emphasis on localization and practical application.

My Mission:

Empowering Localized AI:

To provide intelligent solutions tailored to Chinese-speaking users and local business needs, ensuring that AI tools are not just globally advanced but also culturally and linguistically relevant.

Solving Real-World Problems:

To help businesses and individuals across industries (e.g

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. L27N8750+1💤	0.81	“畜”“烟”“牛肉”
2. L27N6500+1💤	0.80	“番”“米的”“荧”
3. L30N721+1🕒	0.80	“子”“席”“理”
4. L31N4653+1💥	0.80	“院”“顾”“轮”
5. L2N2559-1💥	0.79	“拉”“纳”“卡”
6. L18N605-1💥	0.79	“金”“重”“签”
7. L28N2764+1💤	0.79	“狼”“中心”“宁”
8. L27N8755-1💤	0.79	“istle”“赫”“小人”
9. L31N1613-1💥	0.79	“滤”“嫦”“爛”
10. L30N473-1🕒	0.79	“村镇”“寨”“疡”

Shadow specimen, apart from the ten: L26N1768+1🕒, 0.02.

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 26, neuron 1768, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.66, top-1 agreement 0.72) 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
L26N1768+1🕒 shadow specimen t0-v0.1	0.807	0.65	0.804	1.52	“ assign”“ dere”“ Assign”“assign”“ Assignment”“ Dere”“Assignment”“tern”
L26N1768-1 t0-v0.1	0.657	0.72	0.612	1.29	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.