INDEX

Explanations

understanding why

The neuron is selectively activated by longer, lower‐frequency content words (e.g. unaccustomed, uninterested, Malaysian, travelling).

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

”。

1.23

}.

1.22

”.

1.16

.");

1.16

」。

1.11

។

1.10

}$.

1.09

。

1.09

`.

1.09

 ذریع

1.08

POSITIVE LOGITS

 could

1.41

are

1.39

 were

1.36

 would

1.36

 adalah

1.36

 will

1.31

 ayaa

1.30

was

1.29

may

1.29

などは

1.29

Activations Density 0.445%