INDEX

Explanations

The neuron is specifically triggered by code‐style assignment lines that include string literals (i.e. an equals sign followed by quoted text).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 यही

0.90

㑹

0.85

 stockbild

0.85

Հ

0.78

Ба

0.75

》《

0.75

 سلاټ

0.74

Мо

0.72

性质

0.71

」「

0.70

POSITIVE LOGITS

↵↵

2.22

↵↵↵

2.11

↵↵↵↵

1.91

↵↵↵↵↵

1.54

↵↵↵↵↵↵

1.41

↵↵↵↵↵↵↵

1.22

↵↵↵↵↵↵↵↵

1.19

↵↵↵↵↵↵↵↵↵↵

1.14

↵↵↵↵↵↵↵↵↵

1.06

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.05

Activations Density 0.020%