INDEX

Explanations

observe, study, test, compare

The neuron fires on tutorial‐style, imperative/testing verbs and interrogation words (e.g. “try,” “see,” “if,” “what”) used in code‐example instructions.

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

-1.71

 zwanzig

-1.47

 Aufenthal

-1.34

 vermutlich

-1.34

 bevorzug

-1.30

 Öffnungszeiten

-1.30

 estos

-1.29

 একটি

-1.27

 IUser

-1.25

POSITIVE LOGITS

ถ้า

1.38

cestry

1.36

嚣

1.30

🫁

1.29

涐

1.27

 dicono

1.26

Cependant

1.24

çou

1.21

 kebakaran

1.21

roep

1.20

Activations Density 0.094%