INDEX

Explanations

The neuron fires most strongly on standalone “labels” immediately followed by a colon—i.e. the keys in key:value pairs (like “TMS,” “Email,” “username,” etc.).

amazingly, ridiculously, unbelievably

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

on

-1.55

 garner

-1.36

 epitom

-1.35

for

-1.34

 take

-1.32

out

-1.31

 Generally

-1.26

one

-1.25

 just

-1.23

 changed

-1.20

POSITIVE LOGITS

 ridiculously

1.45

 insanely

1.35

 delightfully

1.32

 sooo

1.30

 amazingly

1.30

 wonderfully

1.26

 vijf

1.25

 magnificent

1.24

 ludicrous

1.23

 unbelievably

1.23

Activations Density 0.000%