© 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 · 2586 ｜ Neuronpedia

Home
GPT2-Small
Transcoders Residuals
8-TRES-DC
2586

INDEX

Explanations

personal reflections or statements beginning with "I."

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

 characterization

-0.68

 incorporation

-0.62

ogh

-0.61

icion

-0.59

姫

-0.59

arthed

-0.57

 assertions

-0.57

ミ

-0.57

INC

-0.57

ゼ

-0.57

POSITIVE LOGITS

 wish

0.86

 wanna

0.85

'm

0.85

 prayed

0.80

 wished

0.79

 begged

0.78

 bombed

0.77

 think

0.76

'd

0.76

've

0.76

Activations Density 0.240%

No Known Activations