INDEX

Explanations

foreign language words and punctuation

This neuron is essentially a “length” detector—it activates most strongly at the ends of long text spans (paragraphs), with activation roughly increasing as the span grows.

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

 Then

-1.96

 what

-1.90

if

-1.86

 When

-1.82

 тогда

-1.80

our

-1.79

 дуже

-1.70

we

-1.67

 when

-1.64

 then

-1.63

POSITIVE LOGITS

 cantora

1.98

 atriz

1.94

 fijne

1.86

还未

1.84

 duur

1.63

 ifølge

1.63

🪛

1.59

 continuación

1.58

 fijn

1.57

 børn

1.56

Activations Density 0.096%