INDEX

Explanations

a or the word after

The neuron fires almost exclusively on the final token of each document segment, i.e. it detects segment-ending words.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

ৃষ্টি

0.74

 personnalisé

0.74

 einen

0.71

 Mechanism

0.71

 Mechanisms

0.71

rer

0.70

നിന്ന്

0.69

ensation

0.69

 graines

0.68

ристо

0.68

POSITIVE LOGITS

sufficiency

0.75

 mediast

0.74

 "/")

0.73

Khokhlov

0.72

SignUp

0.72

>):

0.71

SQ

0.71

ReadAll

0.71

 ώστε

0.71

 prothorax

0.70

Activations Density 0.005%