INDEX

Explanations

plotting and data structures

Explanation of neuron 4 behavior: the main thing this neuron does is find numeric literals (tokens containing digits) in code.

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

if

-1.39

 communication

-1.25

 from

-1.20

if

-1.16

 case

-1.07

-1.06

 after

-1.04

 black

-1.04

 house

-0.96

 communications

-0.96

POSITIVE LOGITS

 garant

1.11

PRODUC

1.02

 Fence

1.01

 estekak

1.00

 guarn

0.99

 rhum

0.99

抱歉

0.98

 conftest

0.98

 lading

0.97

we

0.97

Activations Density 0.007%