INDEX

Explanations

asked to create or do something

The neuron fires most strongly on the principal task‐ or action‐words (usually programming verbs like “create,” “build,” “use,” “write,” etc.) that state what the asker is trying to do.

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

you

-1.98

we

-1.39

 jego

-1.34

 better

-1.30

и

-1.20

 होगा

-1.20

реса

-1.14

can

-1.12

 bemer

-1.12

 और

-1.08

POSITIVE LOGITS

by

1.88

1.49

 лишь

1.42

only

1.35

 because

1.28

Наталья

1.27

 زیرا

1.24

 slechts

1.19

 when

1.18

追求

1.16

Activations Density 0.115%