INDEX

Explanations

street closures

This neuron is acting as a positional counter – it fires more strongly the farther a token is into the chunk, effectively tracking token position in the segment rather than any lexical content.

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

ン

-1.27

 where

-1.26

葑

-1.23

 realizará

-1.21

elfth

-1.20

 consultar

-1.17

 contribution

-1.16

 reciben

-1.16

 nơi

-1.15

ників

-1.15

POSITIVE LOGITS

 mediados

1.40

女

1.38

 cantora

1.34

 atriz

1.34

旖

1.30

chenden

1.28

喜び

1.27

capulco

1.26

 pesar

1.23

 особы

1.22

Activations Density 0.020%