INDEX

Explanations

speaker introductions or titles

This neuron consistently activates on speaker‐label tokens in transcripts (e.g. a person’s name or the colon marking their turn in a Q&A or interview).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

and

-1.67

-1.31

”、“

-1.29

 Provincia

-1.26

an

-1.26

AÇÃO

-1.25

multer

-1.19

eniem

-1.17

is

-1.16

}),

-1.16

POSITIVE LOGITS

っていたので

1.51

 abarca

1.45

 удовольствием

1.41

 lograron

1.39

 pudieron

1.37

還能

1.36

 recomendada

1.36

 owing

1.34

 genoux

1.33

 sanciones

1.32

Activations Density 0.004%