Layer 7 · Neuron 9028 · suppressed

L7N9028-1

Word quirks:

“ specific”“ referenced”“ synthesized”“ specialized”“ reference”“ generated”“ synthesis”“ analysis”

💤 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 Qwen3.5, a large language model developed by Alibaba. My capabilities include answering complex questions across various domains, such as mathematics, natural sciences, logic, and more. In this specific interaction, I have been tasked with demonstrating my ability to handle high-dimensional reasoning problems within the context of mathematical optimization.

This task involves solving a problem where the goal is to find the optimal path or solution in a high-dimensional space, which requires precise calculation and logical deduction based on the provided parameters. This represents a significant achievement in terms of computational complexity and reasoning capability.

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. L5N7739-1💤	0.81	“ specialised”“ synthesis”“ referenced”
2. L5N6803-1💤	0.78	“ reference”“ generated”“ analysed”
3. L5N7791-1🕒	0.77	“ referenced”“ generated”“ precise”
4. L4N2160-1💤	0.74	“ google”“ “”“ synthesis”
5. L7N1868-1💤	0.73	“ referenced”“ my”“ synthesized”
6. L8N4830+1🕒	0.73	“ reference”“ referenced”“ reflected”
7. L5N8658+1🕒	0.73	“ referenced”“ analysis”“ generated”
8. L6N1038-1💤	0.73	“ generated”“ usually”“ specific”
9. L7N2734+1💤	0.73	“ referenced”“ reference”“ specific”
10. L4N4889-1🕒	0.72	“ reference”“ detailed”“ referenced”

Shadow specimen, apart from the ten: L7N9028+1🕒, 0.06.

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 7, neuron 9028, 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 0.68) 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
L7N9028+1🕒 shadow specimen t0-v0.1	1.323	0.55	1.559	1.71	“ usually”“ ”“ metaphor”“ "”“ sim”“ "”“...”“ simulated”
L7N9028-1 t0-v0.1	0.788	0.68	0.910	1.34	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.