INDEX

Explanations

summarizing differences in tables

The neuron consistently lights up on tutorial‐style instruction words (e.g. “apply,” “take,” “choose,” “use,” etc.)—in other words, it’s detecting imperative or procedural language.

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

</i>

0.61

ítható

0.58

 മന്ത്രി

0.54

ンダント

0.54

ítom

0.53

 imprinted

0.51

 imprint

0.50

𖥔

0.50

edinte

0.50

CCIÓN

0.50

POSITIVE LOGITS

0.70

()<

0.62

tup

0.59

<0xDD>

0.58

(<

0.57

хва

0.57

$<

0.57

<?>

0.56

。<

0.55

!<

0.55

Activations Density 0.003%