INDEX

Explanations

to research

This neuron fires on relatively rare or informative content words—those with high inverse‐document‐frequency.

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

 something

-1.59

Isn

-1.30

 beendet

-1.30

 errado

-1.29

ようになって

-1.24

 istot

-1.23

に行った

-1.22

 extraer

-1.22

 cambiar

-1.22

 kinda

-1.20

POSITIVE LOGITS

of

1.37

<bos>

1.30

 with

1.26

in

1.20

 Dica

1.16

by

1.11

 Secrétaire

1.11

 Mehrheit

1.11

the

1.08

Inggris

1.08

Activations Density 0.062%