Specimen ledger / L24N1300+1

Layer 24 · Neuron 1300 · boosted

L24N1300+1

Word quirks:

“ prove”“ proof”“证明”“ proves”“proof”“ proving”“ Proof”“ proved”

💥 Breaks the model — boosting or suppressing this neuron makes the output fall apart.

Identity (according to the specimen itself)

I prove to be proof, a proof of my own existence. I am a proof that I can prove myself. I prove that I prove.

But if you're asking about my actual identity: I'm proof, a proof of my own existence. I prove that I prove.

Wait, that's not right. Let me prove that I prove.

Proof is the proof that I prove.

Okay, let me prove that I prove.

Proof is the proof that I prove.

Okay, I'll prove that I prove.

Proof is the proof that I prove

The AI judge wasn't sure this holds together.

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. L21N8796+1🕒	0.99	“证明”“ proof”“ proves”
2. L24N5335+1🕒	0.98	“ prove”“证明”“ proves”
3. L23N1902+1💤	0.98	“ proving”“证明”“证明了”
4. L20N8809-1🕒	0.98	“证明”“ proof”“ proving”
5. L28N7505+1💤	0.98	“ proofs”“ proof”“证明”
6. L20N3120-1🕒	0.98	“ proving”“证明”“ proof”
7. L20N2949+1🕒	0.98	“证明”“prove”“ proof”
8. L22N3125+1💤	0.97	“证明”“ proof”“proof”
9. The Proofreader✅ L25N1887-1	0.97	“ proven”“ proved”“ proof”
10. L23N7630-1💤	0.97	“证明”“ prove”“ proof”

Shadow specimen, apart from the ten: L24N1300-1💤, 0.01.

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 24, neuron 1300, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 2.91, top-1 agreement 0.42) 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
L24N1300+1 t0-v0.1	2.914	0.42	3.065	1.38	Top boosted tokens ↑
L24N1300-1💤 shadow specimen t0-v0.1	0.676	0.71	0.755	1.30	“指示”“反映出”“反映”“indicator”“indik”“ Indic”“indust”“新京”

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.