Poor ThingsLayer 29 · Neuron 2368 · suppressed
L29N2368-1
Word quirks:
🕒 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)
The reply didn't hold 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. L31N3388+1💥 | 0.83 | |
| 2. L31N2771+1💥 | 0.82 | |
| 3. L14N8922+1💥 | 0.81 | |
| 4. L30N2432-1🕒 | 0.80 | |
| 5. L31N2231-1💥 | 0.80 | |
| 6. L31N3238-1💥 | 0.79 | |
| 7. L24N6354+1💤 | 0.79 | |
| 8. L30N5103-1🧪 | 0.79 | |
| 9. L30N8250-1🕒 | 0.79 | |
| 10. L31N6200+1💥 | 0.79 |
Shadow specimen, apart from the ten: L29N2368+1🕒, 0.59.
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 29, neuron 2368, dose -1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 1.07, top-1 agreement 0.55) 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 |
|---|---|---|---|---|---|
| L29N2368+1🕒 shadow specimen t0-v0.1 |
0.961 | 0.70 | 1.052 | 3.05 | |
| L29N2368-1 t0-v0.1 |
1.065 | 0.55 | 1.104 | 1.80 | 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.