INDEX

Explanations

surprisingly

The neuron strongly activates on words framing rhetorical questions or surprised exclamations (e.g. “How,” “can,” “lead,” “I can’t believe”), i.e. when the text poses incredulous or questioning phrases.

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

 will

0.58

 inevitable

0.57

khidmat

0.57

 interesantes

0.56

にとっては

0.56

 için

0.55

 थोड़ा

0.55

 थोड़ा

0.55

 operation

0.55

 acknowledge

0.53

POSITIVE LOGITS

這麼

1.32

 столь

1.27

 इतनी

1.24

こんな

1.23

如此

1.22

这么

1.17

竟然

1.16

 настолько

1.15

 इतने

1.14

居然

1.12

Activations Density 0.355%