INDEX

Explanations

i will, i do, i was

The neuron fires on first-person expressions (e.g. “I,” “we,” and their auxiliary verbs), marking direct self-reference or speaker commentary.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

our

-1.30

Sådan

-1.29

its

-1.25

Jeden

-1.23

one

-1.20

 mergul

-1.20

 हमारे

-1.17

}+

-1.16

 Kühlschrank

-1.15

其实是

-1.14

POSITIVE LOGITS

riezmann

1.45

']='

1.30

 deberán

1.28

<>("

1.23

 podrán

1.23

bulkan

1.22

</h1>

1.20

 заболеваний

1.20

':'

1.19

ünster

1.19

Activations Density 0.103%