Layer 23 · Neuron 2912 · suppressed

L23N2912-1

Word quirks:

“heck”“adurch”“洞”“ nostr”“殊”“otti”“raj”“stor”

💤 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 AI 3.0, a next-generation AI model designed to provide comprehensive, accurate, and insightful responses across a wide range of topics. My core mission is to empower users by delivering high-quality information, creative support, and analytical insights while ensuring safety, clarity, and adaptability in all interactions.

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. L26N3119+1💤	0.64	“ัด”“烈”“/star”
2. L28N7814-1💤	0.63	“静的”“heim”“尼克”
3. L26N6285+1💤	0.63	“署”“嚏”“ presto”
4. L31N2061-1💥	0.62	“atha”“enio”“熊”
5. L23N665+1💤	0.62	“尤”“ darm”“ vare”
6. L27N4690+1💤	0.62	“oun”“เส”“DUCT”
7. L27N4480+1💤	0.61	“ pann”“轩”“orrh”
8. L26N5442-1💤	0.61	“pack”“ớt”“虞”
9. L22N2029+1💤	0.61	“ensee”“括”“oten”
10. L26N4756+1🕒	0.61	“炮弹”“haus”“窦”

Shadow specimen, apart from the ten: L23N2912+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 23, neuron 2912, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.70, 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
L23N2912+1🕒 shadow specimen t0-v0.1	1.685	0.47	1.748	1.27	“ portion”“ portions”“ parte”“ part”“ parts”“portion”“的那”“part”
L23N2912-1 t0-v0.1	0.699	0.68	0.760	1.37	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.