Layer 11 · Neuron 2608 · suppressed

L11N2608-1

Word quirks:

“ assessment”“ evaluation”“ based”“ assessing”“ directly”“ evaluating”“ evaluations”“ evaluated”

💤 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 Cloud. My mission is to provide high-quality, accurate, and helpful responses to user queries, especially in scenarios where users might have specific needs or expectations. I aim to assist users by offering relevant information, engaging in meaningful conversations, and ensuring that my capabilities align with the user's requirements.

In terms of my identity, I am an AI assistant designed to support users in various tasks, from answering questions to facilitating interactions. My mission is to ensure that I can effectively contribute to user satisfaction, particularly when users require assistance with complex tasks or need reliable information.

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. L10N8419+1🕒	0.96	“ evaluations”“ evaluation”“ assessments”
2. L17N8728+1💤	0.94	“ evaluation”“ evaluating”“ assessment”
3. L16N8799+1🕒	0.94	“ evaluation”“ assessment”“ evaluations”
4. L8N1634-1🕒	0.94	“ evaluation”“ assessment”“ requests”
5. L11N5983-1🕒	0.94	“ evaluation”“ assessment”“ evaluations”
6. L14N5127+1🕒	0.94	“ evaluation”“ evaluator”“ evaluating”
7. L13N7596-1🕒	0.93	“ evaluation”“ evaluations”“ evaluating”
8. L17N7266-1💤	0.93	“ evaluation”“ evaluations”“ evaluate”
9. L10N8718-1🕒	0.93	“ judgement”“ assessment”“ evaluation”
10. L11N3123+1🕒	0.92	“ report”“ presently”“ assessed”

Shadow specimen, apart from the ten: L11N2608+1🕒, 0.16.

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 2608, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.57, 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
L11N2608+1🕒 shadow specimen t0-v0.1	0.965	0.64	1.206	1.36	“ reference”“ because”“ referenced”“ whenever”“ focus”“ here”“ now”“ available”
L11N2608-1 t0-v0.1	0.574	0.72	0.815	1.35	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.