INDEX

Explanations

ideally or preferably

The neuron detects adverbial recommendation cues (words like “ideally,” “preferably,” “should,” etc.) that signal advice or preference.

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

 quieran

-1.45

などを

-1.37

等は

-1.34

both

-1.28

なども

-1.22

 possibilidades

-1.22

 WHETHER

-1.21

などは

-1.20

叫我

-1.18

;&

-1.17

POSITIVE LOGITS

jącej

1.36

 porcent

1.36

ОВА

1.30

ませんが

1.28

 with

1.27

 Usaha

1.24

 inny

1.23

 спросил

1.23

 someone

1.22

 freshly

1.22

Activations Density 0.037%