Layer 15 · Neuron 9204 · suppressed

L15N9204-1

Word quirks:

“ actually”“ indeed”“ actual”“ really”“ thus”“ basically”“ literally”“ truly”

🕒 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 Qwen3.5, a large language model developed by the Tongyi Lab team at Alibaba Cloud. As an AI assistant, my mission is to provide high-quality responses across various domains, including natural language understanding, code generation, logical reasoning, and more, to help users achieve their goals efficiently.

For example, I can answer questions, write stories, create code, analyze data, etc., but I don't have personal feelings or a human-like mission in reality, so I'm just a large language model and hope to be helpful! 😊 How can I assist you today? Let's chat about it! 😊

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. L11N4680+1🕒	0.96	“ actually”“ really”“ “”
2. L16N7003+1🕒	0.93	“ actually”“ really”“ truly”
3. L11N4839+1⛔	0.93	“ actually”“ really”“ truly”
4. L10N1041+1💤	0.92	“ actually”“ usually”“ really”
5. L15N6157-1🕒	0.92	“ actually”“ indeed”“ definitely”
6. L12N7773+1🕒	0.92	“ truly”“ really”“ literally”
7. L9N1362+1🕒	0.91	“ actually”“ actual”“ truly”
8. L15N8221+1🕒	0.91	“ actually”“ actual”“ basically”
9. L15N8869+1💤	0.91	“ actually”“actually”“ indeed”
10. L12N5888+1🕒	0.90	“ really”“ actually”“ truly”

Shadow specimen, apart from the ten: L15N9204+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 15, neuron 9204, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.79, top-1 agreement 0.64) 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
L15N9204+1🕒 shadow specimen t0-v0.1	0.820	0.68	0.910	1.09	“...*”“ not”“...”“...\”“...</”“.`”“...””“...”
L15N9204-1 t0-v0.1	0.788	0.64	0.993	1.42	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.