INDEX

Explanations

code comments and identifiers

np_acts-logits-general · gemini-2.5-flash-lite

whitespace at the beginning of lines, particularly in code or structured text documents.

oai_token-act-pair · claude-4-5-sonnet Triggered by @bfgcn7frf5

The neuron consistently lights up on lines that look like section headers or all-caps log/metadata titles (e.g. “PARSING FILE …”, “BEERS:”), i.e. uppercase headings.

oai_token-act-pair · o4-mini Triggered by @jyhe0408

New Auto-Interp

Configuration

google/gemma-scope-27b-pt-res/layer_10/width_131k

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

if

-2.36

 itself

-1.79

or

-1.75

🌸

-1.62

was

-1.55

ophageal

-1.51

 tica

-1.51

 epitom

-1.49

 three

-1.48

it

-1.48

POSITIVE LOGITS

are

2.64

 themselves

2.23

 которые

2.08

pecabe

1.95

 których

1.93

玙

1.81

nuevos

1.77

 هستند

1.74

 trám

1.74

 которыми

1.73

Activations Density 0.126%

code comments and identifiers

whitespace at the beginning of lines, particularly in code or structured text documents.

The neuron consistently lights up on lines that look like section headers or all-caps log/metadata titles (e.g. “PARSING FILE …”, “BEERS:”), i.e. uppercase headings.

No Comments

No Known Activations

code comments and identifiers

whitespace at the beginning of lines, particularly in code or structured text documents.

The neuron consistently lights up on lines that look like section headers or all-caps log/metadata titles (e.g. “PARSING FILE …”, “BEERS:”), i.e. uppercase headings.

No Comments

No Known Activations