INDEX

Explanations

actions and responses

The neuron is highly responsive to the standalone term “action” (including its title‐case or plural “Actions”) in the text.

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

-1.64

 dorp

-1.33

 gardes

-1.30

of

-1.29

if

-1.27

-1.23

 bleu

-1.22

笈

-1.21

ent

-1.21

POSITIVE LOGITS

 quedará

1.55

 ontem

1.47

 Verhalten

1.41

偸

1.41

 comenzará

1.41

 portatil

1.40

 Handeln

1.39

 Encuentra

1.37

 dará

1.36

 hará

1.35

Activations Density 0.037%