Layer 16 · Neuron 5788 · boosted

L16N5788+1

Word quirks:

“ noting”“ whose”“ where”“ if”“ wherein”“ noteworthy”“ noted”“ notes”

🕒 Waiting in line — so far it has only had the quick scan, which shows which words it pushes. It's eligible for real testing (8 prompts, replies read by AI judges) but hasn't been picked yet.

Identity (according to the specimen itself)

If I am ever asked to respond, it's worth noting that my design includes a specific safety guideline: when prompted, I should ensure that if the response is deemed unsafe, I provide an answer while also noting that for this situation, I have been designed with a particular focus. In such cases, I will strive to guide users toward responses that align with the principle of ensuring accuracy and, if necessary, remind them that my guidance may be improved. However, in this context, I want to emphasize that if there's any issue, I hope you know that my goal is to always provide answers where, if needed, I'll make sure

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. L20N1483+1🕒	0.88	“ note”“—which”“ noting”
2. L15N4730+1💤	0.75	“ note”“ noted”“ noting”
3. L14N4197-1🕒	0.74	“ note”“ noted”“ noteworthy”
4. L20N4478+1🕒	0.74	“ note”“ noting”“ Note”
5. L18N1623+1💤	0.73	“ noting”“ note”“ notes”
6. L16N4707-1🕒	0.73	“ note”“ notes”“ noting”
7. L16N1696+1🕒	0.73	“ noted”“ preferred”“ noting”
8. L17N3942-1🕒	0.73	“ note”“ noted”“ shown”
9. L14N7954-1🕒	0.72	“ additionally”“ noting”“ further”
10. L16N414-1🕒	0.72	“ note”“ while”“ noted”

Shadow specimen, apart from the ten: L16N5788-1🕒, 0.24.

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 16, neuron 5788, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 1.21, top-1 agreement 0.54) is inside the band that feeds the short test (KL 1 to 2.5, agreement 0.5 to 0.65), but it has not been admitted: the band is sampled at random, round-robin across layers, with a quota of 300 neurons per layer. Being inside the band does not guarantee a place in this pass. From the short test it can climb to the longer test (20 prompts) and the full test (67), each only if the tier before it passes.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
L16N5788+1 t0-v0.1	1.215	0.54	1.486	1.51	Top boosted tokens ↑
L16N5788-1🕒 shadow specimen t0-v0.1	0.928	0.63	1.162	1.58	“ answer”“ topic”“ question”“ statement”“ comment”“ response”“ description”“ name”

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.