INDEX

Explanations

choosing from or what

The neuron spikes on the one or two words that stand out as the “tag” or title‐like pivot of each segment—i.e. the single token that most strongly signals a new topical section.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

чный

0.98

0.89

 учены

0.85

 ativos

0.83

 этому

0.82

aar

0.82

ỡng

0.82

 тобой

0.80

 verdade

0.80

 خاطر

0.79

POSITIVE LOGITS

0.81

:]

0.74

 legs

0.72

:][

0.71

غط

0.70

דיה

0.70

:])

0.70

פות

0.69

:"))

0.68

קט

0.66

Activations Density 0.000%