INDEX

Explanations

electroshocked into

This neuron fires on words that form common two-word or multi-word collocations (e.g. “toughen up,” “yield point,” “Feux sur,” “how to”).

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

 Matériau

-1.55

 Accessoires

-1.41

 intensify

-1.39

 gethan

-1.39

𝑿

-1.36

 jopa

-1.35

 навіть

-1.34

特に

-1.34

 между

-1.33

 вроде

-1.30

POSITIVE LOGITS

↵↵↵

1.88

to

1.63

 from

1.48

Nuestra

1.46

↵↵↵↵

1.39

1.38

Tenemos

1.35

ircut

1.30

 after

1.27

 during

1.26

Activations Density 0.123%