INDEX

Explanations

can, used, or possible

The neuron fires on auxiliary/modal verbs and related tokens (e.g. English “can,” “be,” “used,” “have,” “could,” “was,” etc. and Russian “Можно,” “ли,” “сказать,” “сделать”), i.e. constructions expressing permission, possibility or passive voice.

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

 these

-1.34

its

-1.28

↵↵↵

-1.27

We

-1.24

we

-1.20

้ย

-1.20

ers

-1.17

.³

-1.14

Hæ

-1.14

POSITIVE LOGITS

缶

1.51

 dispozici

1.43

 memanfaatkan

1.41

がありますが

1.41

 agujas

1.34

 libertà

1.33

 ocen

1.32

 okno

1.28

zni

1.27

 funzionalità

1.27

Activations Density 0.011%