INDEX

Explanations

This neuron is activated by tokens that look like acronyms or uppercase abbreviations (including hashtagged acronyms).

images and aviation

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

for

-2.11

 from

-2.05

 with

-1.96

 eight

-1.85

at

-1.84

on

-1.81

in

-1.78

 their

-1.77

to

-1.72

as

-1.72

POSITIVE LOGITS

 avancer

1.34

 participer

1.31

kowa

1.30

 corriger

1.30



1.29

 görüntüleri

1.28

(\<

1.27

 Dinas

1.26

 horloge

1.26

 inder

1.24

Activations Density 0.001%