INDEX

Explanations

list of mathematical indices or variable names

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

phrases that convey superlatives or high-degree evaluations indicating exceptional prominence, importance, or extremity.

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

the neuron detects capitalized named entities and title-case noun phrases (proper names, product names, and formal job/role titles).

oai_token-act-pair · gpt-5-mini Triggered by @smilgaveronika

customer service and support related content.

oai_token-act-pair · gpt-5-nano Triggered by @smilgaveronika

The neuron strongly activates on multiword title‐cased phrases used in formal or marketing contexts (e.g. product names, corporate titles, branded jargon).

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

New Auto-Interp

Configuration

google/gemma-scope-2-1b-pt/resid_post/layer_13_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

 đều

1.79

 kanë

1.49

Are

1.48

Were

1.44

są

1.44

 janë

1.41

θούν

1.38

 всі

1.36

 fanno

1.34

ቸውን

1.33

POSITIVE LOGITS

 gaat

1.45

has

1.42

 houdt

1.38

 wordt

1.37

 heeft

1.35

is

1.35

 werd

1.35

 wird

1.33

 решил

1.30

kills

1.30

Activations Density 0.250%

list of mathematical indices or variable names

phrases that convey superlatives or high-degree evaluations indicating exceptional prominence, importance, or extremity.

the neuron detects capitalized named entities and title-case noun phrases (proper names, product names, and formal job/role titles).

customer service and support related content.

The neuron strongly activates on multiword title‐cased phrases used in formal or marketing contexts (e.g. product names, corporate titles, branded jargon).

No Comments

No Known Activations

list of mathematical indices or variable names

phrases that convey superlatives or high-degree evaluations indicating exceptional prominence, importance, or extremity.

the neuron detects capitalized named entities and title-case noun phrases (proper names, product names, and formal job/role titles).

customer service and support related content.

The neuron strongly activates on multiword title‐cased phrases used in formal or marketing contexts (e.g. product names, corporate titles, branded jargon).

No Comments

No Known Activations