INDEX

Explanations

academic paper introductions

The neuron strongly activates on structural metadata and section‐heading tokens (e.g. author initials with periods, email “@” signs, “Abstract,” “Introduction,” affiliations), i.e. it detects document headers and metadata markers.

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

-1.94

-1.86

-1.85

-1.82

-1.77

-1.72

“

-1.68

↵

-1.68

-1.58

POSITIVE LOGITS

 Однако

2.02

 peraturan

1.89

 although

1.80

出来て

1.66

Although

1.59

 after

1.58

 despite

1.58

although

1.55

㍑

1.53

 meskipun

1.52

Activations Density 0.046%