INDEX

Explanations

file paths and code snippets

The neuron fires strongly on the final segment of multi‐part names or identifiers (i.e. the trailing token in a compound word, file path, or author name).

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

and

-2.39

or

-2.20

 both

-1.96

 without

-1.95

 before

-1.91

 with

-1.86

not

-1.86

for

-1.83

 Before

-1.74

at

-1.73

POSITIVE LOGITS



2.11

킁

1.63

玊

1.59

 Fasting

1.59

媄

1.55

鹋

1.55

槼

1.54

沵

1.52

 evocative

1.52

 patate

1.52

Activations Density 0.062%