INDEX

Explanations

numbers, punctuation, and common words

This neuron detects sentence-level discourse markers or transitional adverbs (e.g. “though,” “either,” “amongst,” “respectively,” often appearing with surrounding punctuation)

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

𝚝

0.91

м

0.85

чей

0.85

مپ

0.81

𝚑

0.79

 equilibration

0.78

ровали

0.77

otherArchive

0.77

রাজনৈতিক

0.75

्रीम

0.75

POSITIVE LOGITS

是

0.91

은

0.84

 than

0.82

带

0.82

选择

0.81

0.78

輛

0.78

，

0.77

、

0.76

则

0.76

Activations Density 0.101%