INDEX

Explanations

transition words

The neuron fires selectively on the first one or two tokens of new sentences—especially common discourse markers or function words (e.g. “But,” “In fact,” “Whether,” “And,” “So,” “Oh”)—i.e. it detects sentence-initial cues.

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

 délais

-1.77

 ซึ่ง

-1.73

 réinitial

-1.73

 extremely

-1.72

 estadounid

-1.70

 déchets

-1.69

鵙

-1.68

튿

-1.66

傒

-1.66

 étan

-1.66

POSITIVE LOGITS

 with

2.14

for

1.94

at

1.83

as

1.78

not

1.73

 there

1.70

or

1.66

 after

1.63

 without

1.61

 some

1.60

Activations Density 0.006%