INDEX

Explanations

Are followed by tai or pa

This neuron fires on the first content token of each text snippet or heading, i.e. it detects segment‐start words.

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

porque

-1.38

Karena

-1.35

Setiap

-1.33

Tujuan

-1.33

írez

-1.30

𝓦

-1.30

Neden

-1.28

Pemer

-1.27

 纹身

-1.25

 sécher

-1.25

POSITIVE LOGITS

1.73

1.55

🙁

1.44

ichtete

1.38

to

1.37

MAXN

1.36

)–

1.34

**)

1.31

 sich

1.30

beforeAll

1.30

Activations Density 0.224%