INDEX

Explanations

unbiased, neutral, impartial

The neuron is strongly tracking a token’s position in the document (i.e. later tokens get higher activations).

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

initely

-0.78

vip

-0.78

<\

-0.75

ducers

-0.75

生きて

-0.74

MEMOR

-0.72

|'('

-0.72

༅

-0.71

individ

-0.71

انیا

-0.71

POSITIVE LOGITS

 unbiased

4.34

 neutral

4.34

 impartial

4.06

 objectivity

3.75

neutral

3.70

 objective

3.67

 neutrality

3.61

 Neutral

3.47

Neutral

3.47

 impartiality

3.36

Activations Density 0.093%