The neuron strongly activates on standalone numeric tokens (e.g. dates or numbers in the text).

dates in July

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

-1.84

-1.52

-1.49

-1.48

 different

-1.46

en

-1.46

-1.44

In

-1.43

-1.41

POSITIVE LOGITS

 servis

1.59

 norvég

1.56

 COMPOUNDS

1.54

 évé

1.53

είς

1.51

 recommande

1.50

 linge

1.50

 situe

1.49

asties

1.49

yourself

1.48

Activations Density 0.045%