INDEX

Explanations

couple of, through, chatbot, house lease, decided on

This neuron doesn’t pick out any coherent linguistic pattern—instead it only lights up on isolated, seemingly random or malformed tokens (essentially acting as a “garbage” detector rather than detecting a real word category).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

0.97

0.92

ਆਂ

0.86

KYO

0.86

ıları

0.81

0.80

ل

0.79

с

0.78

م

0.78

kick

0.77

POSITIVE LOGITS

 deformations

0.96

 quím

0.81

 beasts

0.79

諡

0.77

éc

0.76

rechts

0.76

 lotions

0.76

 esetén

0.75

 lekin

0.75

 detonations

0.75

Activations Density 0.001%