INDEX

Explanations

selectively view / class A / way to

The neuron fires on high-register, Latinate content words (often abstract nouns or adverbs, e.g. “complicity,” “selectively,” proper names) rather than on common function words.

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

 even

-1.31

and

-1.30

 although

-1.27

 only

-1.27

 still

-1.24

 nothing

-1.23

<bos>

-1.15

-1.12

 stronę

-1.10

erobic

-1.10

POSITIVE LOGITS

viewDid

1.22

оны

1.20

 basadas

1.16

 francès

1.14

்ட

1.13

Liet

1.12

葬

1.11

afficher

1.09

orderid

1.09

starttime

1.06

Activations Density 0.679%