INDEX

Explanations

This neuron isn’t picking out any semantic pattern in the words themselves but simply fires in proportion to a token’s position in the text (early tokens get a higher activation, later ones lower).

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

words related to continuation or ongoing processes, particularly verbs ending in "-ing" that denote activities being maintained or carried forward.

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

verbs describing actions or states, especially continuous or passive forms ending in -ing or -ed.

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

New Auto-Interp

Configuration

google/gemma-scope-2-12b-pt/resid_post/layer_24_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

しかも

0.77

ga

0.75

Cela

0.73

gap

0.73

 लाइक

0.71

Eg

0.71

Gap

0.68

Says

0.68

their

0.67

 Deshalb

0.67

POSITIVE LOGITS

не

0.90

 ylabel

0.82

റൽ

0.78

 disintegrated

0.78

adj

0.77

อร์

0.76

 massacre

0.75

 transm

0.74

ʛ

0.74

 crusade

0.74

Activations Density 0.000%

This neuron isn’t picking out any semantic pattern in the words themselves but simply fires in proportion to a token’s position in the text (early tokens get a higher activation, later ones lower).

words related to continuation or ongoing processes, particularly verbs ending in "-ing" that denote activities being maintained or carried forward.

verbs describing actions or states, especially continuous or passive forms ending in -ing or -ed.

No Comments

No Known Activations

This neuron isn’t picking out any semantic pattern in the words themselves but simply fires in proportion to a token’s position in the text (early tokens get a higher activation, later ones lower).

words related to continuation or ongoing processes, particularly verbs ending in "-ing" that denote activities being maintained or carried forward.

verbs describing actions or states, especially continuous or passive forms ending in -ing or -ed.

No Comments

No Known Activations