INDEX

Explanations

code formatting and punctuation

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

opening brackets in structured text or code.

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

The neuron is selectively activating on metadata tokens—things like headings, titles, author names, URLs/domains, and other proper‐noun or label‐style words.

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

-1.79

-1.59

</

-1.55

 this

-1.49

this

-1.48

Despite

-1.47

-1.44

糞

-1.42

our

-1.40

The

-1.39

POSITIVE LOGITS

 ongeveer

1.56

 beberapa

1.55

alties

1.45

以外にも

1.40

スカート

1.40

 Anſ

1.38

 tujuh

1.38

两者

1.37

댔

1.36

撄

1.35

Activations Density 0.047%

code formatting and punctuation

opening brackets in structured text or code.

The neuron is selectively activating on metadata tokens—things like headings, titles, author names, URLs/domains, and other proper‐noun or label‐style words.

No Comments

No Known Activations

code formatting and punctuation

opening brackets in structured text or code.

The neuron is selectively activating on metadata tokens—things like headings, titles, author names, URLs/domains, and other proper‐noun or label‐style words.

No Comments

No Known Activations