Layer 11 · Neuron 6659 · boosted

L11N6659+1

Word quirks:

“ generated”“ generating”“ outputs”“ generation”“ gen”“ output”“ synthesized”“ created”

🕒 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, the latest large language model developed by Alibaba. Here is my mission:

Qwen3.5 Mission:
To provide high-quality responses to user queries by leveraging advanced reasoning capabilities, knowledge retrieval, and multi-modal analysis. This involves generating coherent, accurate, and contextually relevant outputs across a wide range of topics.

Capabilities:

Reasoning: I can perform complex logical reasoning tasks, solve math problems, and analyze data.
Knowledge Retrieval: I can access vast amounts of information to answer questions.
Multi-modal Analysis: I can interpret images

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. L11N3020+1🕒	0.94	“ generated”“ generates”“ output”
2. L5N6393-1🕒	0.92	“ generated”“ output”“ generating”
3. L8N3991+1🕒	0.91	“ generating”“ output”“ generation”
4. L8N3748-1🕒	0.91	“ output”“ generated”“ outputs”
5. L15N8187+1🕒	0.91	“ generated”“ products”“ generating”
6. L6N4698+1💤	0.89	“ generated”“ generating”“ code”
7. L7N7781+1💤	0.89	“ generating”“ generates”“ specific”
8. L7N2007-1💤	0.88	“ generated”“ generating”“ generate”
9. L6N8831-1🕒	0.88	“ generation”“ generated”“ generating”
10. L30N7181+1💥	0.88	“ generated”“ generating”“ Generating”

Shadow specimen, apart from the ten: L11N6659-1🕒, 0.23.

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 11, neuron 6659, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.88, top-1 agreement 0.62) 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
L11N6659+1 t0-v0.1	0.884	0.62	1.112	1.36	Top boosted tokens ↑
L11N6659-1🕒 shadow specimen t0-v0.1	0.936	0.63	1.183	1.44	“ ONLY”“ collecting”“ trying”“ grouping”“ simple”“ groups”“ `”“ only”

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.