INDEX

Explanations

code closing syntax

The neuron strongly activates on long runs of indentation whitespace (i.e. the multi‐space prefixes that mark code block lines).

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

Ħ

-1.20

 bancaria

-1.19

 meni

-1.13

癌症

-1.12

НТ

-1.07

 аналоги

-1.07

 Avez

-1.06

 cokelat

-1.05

 italiana

-1.05

гря

-1.05

POSITIVE LOGITS

Кроме

1.17

前は

1.13

↵↵↵↵↵↵↵↵↵↵↵↵↵

1.10

↵↵↵↵↵↵↵↵↵↵

1.08

ناية

1.07

<bos>

1.06

擰

1.05

Filme

1.05

炜

1.04

{~

1.03

Activations Density 0.004%