INDEX

Explanations

applying to groups or games

The neuron strongly fires on imperative “call-to-action” language—especially form-style instructions (e.g. “fill,” “post,” “go,” “can,” “do”) and field labels directing the reader to take an action.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 when

-1.92

 from

-1.70

 during

-1.58

 after

-1.57

-1.40

んだから

-1.32

 then

-1.29

 information

-1.24

 deceitful

-1.23

 which

-1.22

POSITIVE LOGITS

who

1.49

𓆏

1.45

1.34

%).

1.33

 domy

1.29

1.27

skiego

1.27

1.26

Nigeria

1.24

mostly

1.24

Activations Density 0.163%