INDEX

Explanations

best to do

The neuron detects comparative/superlative recommendation phrases—words like “best,” “easier,” or “easiest” (often paired with “to” or “off”) that signal giving advice or a preferred action.

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

</b>

-2.11

</i>

-1.51

 ());

-1.41

-1.39

いいのか

-1.38

());

-1.36

 quiere

-1.32



-1.31

気軽

-1.31

POSITIVE LOGITS

 murs

1.55

鷙

1.52

to

1.52

 ſont

1.51

 diriger

1.48

หรือ

1.43

 calendriers

1.42

 bricola

1.41

 bascul

1.39

 plafon

1.37

Activations Density 0.023%