© Neuronpedia 2026

Privacy & Terms Blog GitHub Slack Twitter Contact

Neuronpedia

Jacobian LensNEW

Natural Language

NEW Assistant AxisNEW Circuit TracerUPDATESteer SAE Evals Exports interp-engineNEWAPI Community Blog Privacy & Terms Contact

GPT2-SMALL · 8-TRES-DC · 2635 ｜ Neuronpedia

Home
GPT2-Small
Transcoders Residuals
8-TRES-DC
2635

INDEX

Explanations

the conjunction "and" being used in various contexts

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Embeds

Show PlotsShow ExplanationShow ActivationsShow Test FieldShow SteerShow Link

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

dict

-0.77

comb

-0.72

entials

-0.72

ーティ

-0.71

among

-0.71

ゼ

-0.70

amn

-0.67

 Heard

-0.66

OX

-0.65

umeric

-0.65

POSITIVE LOGITS

 instead

0.86

 Instead

0.79

Instead

0.79

instead

0.75

 revert

0.70

Rather

0.70

 Rather

0.67

 rather

0.66

]=

0.64

 adopt

0.63

Activations Density 0.250%

No Known Activations