INDEX

Explanations

difficult to do

The neuron fires strongly on verbs that describe taking action or performing tasks—especially cognitive or workflow actions like “get,” “grasp,” “motivate,” “focus,” “edit,” or “convert.”

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

 instead

-1.48

 !!!!!

-1.25

 .......

-1.20

 вместо

-1.20

!!!!!!!!!

-1.19

preferably

-1.18

Instead

-1.17

 !!!!

-1.16

 !!!!!!

-1.14

 ........

-1.13

POSITIVE LOGITS

 because

2.36

 people

1.69

 even

1.65

due

1.63

 Because

1.61

 debido

1.41

get

1.39

new

1.35

 getting

1.31

 difficult

1.31

Activations Density 0.068%