INDEX

Explanations

as if were

The neuron selectively fires on non‐base (inflected or conjugated) verb forms—e.g. past‐tense, subjunctive, and participle tokens.

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

 there

-1.57

״

-1.35

<eos>

-1.31

pas

-1.29

hè

-1.27

blocks

-1.26

return

-1.25

 "));

-1.24

身后

-1.23

 religi

-1.22

POSITIVE LOGITS

1.77

1.67

 However

1.63

1.56

 zawsze

1.50

 周年

1.45

 alltid

1.44

 всегда

1.44

Médaille

1.44

hastly

1.42

Activations Density 0.174%