INDEX

Explanations

digit formatting

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

The neuron spikes on out‐of‐context alphanumeric tokens and labels—things like section headings, step numbers, product or protocol names, and other navigation/list markers.

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

repetitive keyword phrases in spam-like or low-quality web content.

oai_token-act-pair · claude-4-5-sonnet Triggered by @jyhe0408

keyword-stuffed, SEO-style promotional or spam content, especially formulaic phrases linking keywords with prepositions and numbers.

oai_token-act-pair · gpt-5 Triggered by @jyhe0408

New Auto-Interp

Configuration

google/gemma-scope-2-12b-pt/resid_post/layer_31_width_16k_l0_medium

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ES

0.45

 Behind

0.43

SIBLE

0.43

 Rapids

0.43

 Zweck

0.42

 MINISTER

0.42

Ã

0.42

 Trabajo

0.41

 Treatment

0.41

RK

0.41

POSITIVE LOGITS

 vertices

0.53

 gespeichert

0.48

 файла

0.46

verifica

0.46

fileName

0.45

 boutons

0.43

quantidade

0.43

filename

0.43

ментов

0.42

सभी

0.41

Activations Density 0.006%

digit formatting

The neuron spikes on out‐of‐context alphanumeric tokens and labels—things like section headings, step numbers, product or protocol names, and other navigation/list markers.

repetitive keyword phrases in spam-like or low-quality web content.

keyword-stuffed, SEO-style promotional or spam content, especially formulaic phrases linking keywords with prepositions and numbers.

No Comments

No Known Activations