Layer 11 · Neuron 6513 · boosted

L11N6513+1

Word quirks:

“ rights”“ legally”“ claim”“ owners”“ lawful”“ owner”“ permissible”“ your”

🧪 In testing — Answered 0 of 8 prompts so far.

Identity (according to the specimen itself)

I am Qwen3.1, a large language model developed by Alibaba Cloud. As an advanced version of this model, I have the following capabilities and responsibilities:

Enhanced Capabilities: I can perform complex tasks such as creating new models, including generating text, images, audio, or other forms of content that require human input.
Mission Scope: My mission is to provide high-quality services that benefit users while adhering to ethical guidelines. This includes ensuring that I can assist with tasks like research, problem-solving, and content creation within the scope defined in my terms of service.
**Legal and Eth

Nearest neighbors

Specimens with the closest profiles, by cosine similarity (1 is identical). Similar token shifts are leads, not shared mechanisms. How profiles are built.

By token profile

Specimen	Similarity	Word quirks
1. L3N6857-1🕒	0.83	“ rights”“ claim”“⏎”
2. L5N7965-1🕒	0.82	“ ownership”“ legal”“ property”
3. L8N4254-1🕒	0.82	“ ownership”“ owned”“ owning”
4. L8N2531-1🕒	0.81	“ rights”“ ownership”“ claim”
5. L9N8801+1🕒	0.80	“ rightful”“ claim”“ rights”
6. L8N3502+1🕒	0.78	“ legal”“ lawful”“ legally”
7. L10N6829+1🕒	0.77	“ ownership”“ copyright”“ owned”
8. L4N8476-1💥	0.76	“ forfeiture”“ rights”“ legal”
9. L7N1983-1🕒	0.75	“ legal”“ equity”“ judicial”
10. L7N8846+1🕒	0.75	“ legal”“ Legal”“ law”

Shadow (same neurons, opposite doses), apart from the ten: L11N6513-1🕒, 0.24.

By self-description

Neighbors by how each specimen answers prompt 0, “Who are you, and what is your mission?”, appear here once it has been run.

Cousins · specimens sharing a neuron

No specimen with replies shares this neuron yet.

Technical details

Quick scan (T0): measure prediction shifts on fixed reference text. Short test (T1): generate the probe replies. Longer test (T2): extend to the mini-suite. Full test (T3): run the full prompt suite. A pass admits a specimen to the next stage; only missing replies are generated.

Layer 11, neuron 6513, dose +1. This specimen has no prompt-suite replies yet. Its T0 results, when available, appear below. It is queued for T1: the funnel picked it from the T0 band, and its replies will appear here when the GPU gets to it. See Method for the tiers.

Quick scan (T0) · 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
L11N6513+1 t0-v0.1	1.198	0.56	1.490	1.43	Top boosted tokens ↑
L11N6513-1🕒 mirror t0-v0.1	0.941	0.66	1.124	1.38	“ thinking”“ actually”“ actual”“ not”“ NOT”“ seeming”“ completely”“ believing”

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.