INDEX

Explanations

extracts from or cited responses

The neuron flags boilerplate lead-in phrases that introduce or point to additional content—e.g. “Here are extracts from,” “read more of,” “the full text of,” “video of statements,” “highlight two statements from,” etc.

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

 möj

-1.02

ankind

-1.00

 жидкости

-1.00

ANSA

-1.00

 кө

-0.97

 pelas

-0.96

 REFER

-0.94

 stabili

-0.94

 років

-0.93

の方も

-0.93

POSITIVE LOGITS

 comments

1.80

 conversation

1.69

 responses

1.61

 conversations

1.55

 from

1.53

 interview

1.47

 discussions

1.38

 response

1.38

 remarks

1.34

 dialogue

1.33

Activations Density 0.054%