INDEX

Explanations

Summer Session

The neuron consistently activates on tokens that are parts of capitalized names or titles (i.e. proper nouns).

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

();

-1.46

 horloge

-1.37

’;

-1.36

 kredi

-1.33

schon

-1.33

 artig

-1.32

leicht

-1.30

garanti

-1.30

dö

-1.30

Procedimento

-1.29

POSITIVE LOGITS

</strong>

1.77

</b>

1.63

 with

1.63

 only

1.63

1.40

</em>

1.38

 because

1.38

1.34

 from

1.29

 just

1.29

Activations Density 0.004%