INDEX

Explanations

Names followed by punctuation

The neuron fires on isolated uppercase-letter tokens (often with periods) corresponding to authors’ initials in reference citations.

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

 from

-2.67

who

-2.38

 their

-2.17

by

-1.80

all

-1.77

 also

-1.70

any

-1.68

it

-1.59

 both

-1.58

 this

-1.47

POSITIVE LOGITS

⩴

1.34

 entraî

1.29

graphed

1.28

܆

1.27

 élimin

1.27

菇

1.26

 téléphone

1.24

轤

1.23

沨

1.23

᥀

1.23

Activations Density 0.007%