INDEX

Explanations

handle object

The neuron strongly activates on programming‐style tokens—e.g. function names, uppercase identifiers, acronyms, punctuation, and numeric or code‐like elements—i.e. bits of source code rather than ordinary prose.

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

 이야기를

0.74

 trimestre

0.73

 mús

0.73

 inconsider

0.72

 campaña

0.70

 montaña

0.68

 trés

0.68

 rebranded

0.68

 incrível

0.68

 приносит

0.68

POSITIVE LOGITS

Ptr

1.24

ptr

1.13

ptr

1.05

Handle

1.04

nullptr

0.98

PTR

0.98

 Handle

0.96

 nullptr

0.93

 Pointer

0.89

 destructor

0.89

Activations Density 0.035%