Layer 28 · Neuron 2848 · boosted

L28N2848+1

Word quirks:

“用一个”“在一个”“对于一个”“volve”“从一个”“成了一个”“/a”“ care”

🕒 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)

I'm a large language model developed by a company. A key part of my design is to be helpful, harmless, and a good citizen. I aim to support a wide range of tasks, from a quick answer to a complex project.

A core part of my mission is to be a reliable and safe resource for a global user base. I strive to provide accurate information, respect a diverse range of perspectives, and avoid generating a harmful or a misleading response.

How can I help you today?

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. L25N5583-1🕒	0.88	“在一个”“到一个”“为一个”
2. L23N1544+1🕒	0.82	“从一个”“为一个”“对于一个”
3. L31N4703+1💥	0.82	“一个”“是一个”“的一个”
4. L25N1012-1🕒	0.82	“这是个”“一个”“是一个”
5. L26N583-1🕒	0.81	“一个”“是一个”“而是一个”
6. L29N3119+1🕒	0.80	“一个”“ a”“有一个”
7. L30N796-1🕒	0.80	“ Sebuah”“是一个”“ eine”
8. L22N3669+1💤	0.79	“一个大”“一个小”“一个”
9. L24N7850-1💤	0.79	“有一个”“做一个”“有一個”
10. L29N1121-1💤	0.78	“的一个”“着一个”“ jednym”

Shadow specimen, apart from the ten: L28N2848-1💥, 0.18.

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 28, neuron 2848, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 1.09, top-1 agreement 0.61) 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
L28N2848+1 t0-v0.1	1.088	0.61	1.153	1.77	Top boosted tokens ↑
L28N2848-1💥 shadow specimen t0-v0.1	3.955	0.17	4.067	0.56	“a”“ a”“A”“ A”“an”“对不对”“а”“*A”

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.