INDEX

Explanations

symbols followed by common words

This neuron predominantly activates on code‐style or specially formatted subtokens—things like punctuation marks (e.g. “/”), identifiers with underscores, and code keywords (e.g. “printing”).

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

 همچنین

-0.91

Kesimpulan

-0.87

 épa

-0.84

本当は

-0.83

仕上がり

-0.81

まさか

-0.81

まるで

-0.77

 Dolce

-0.77

 natür

-0.77

Dol

-0.77

POSITIVE LOGITS

 again

1.12

and

1.11

you

1.02

of

0.96

 Again

0.90

<>();

0.88

endell

0.83

ulho

0.82

 lieu

0.82

 raccont

0.81

Activations Density 0.002%