INDEX

Explanations

where he or she

The neuron sharply activates on relative‐clause introduction words (e.g. “which,” “where”), i.e. tokens that signal the start of a subordinate clause.

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

}();

-1.44

 this

-1.38

浹

-1.38

～～～

-1.38

飈

-1.36

Unlike

-1.33

Although

-1.33

 only

-1.30

七十

-1.29

（

-1.27

POSITIVE LOGITS

 fehler

1.41

 タンク

1.38

 schrift

1.33

⑊

1.30

owanie

1.29

 lép

1.29

1.28

est

1.27

 aveva

1.27

Activations Density 0.013%