INDEX

Explanations

adding configuration or elements

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

the word "the" in technical or instructional contexts where it precedes specific items to be added, modified, or referenced.

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

The neuron fires on the imperative “how‐to” instruction phrases—e.g. verbs like “download” or “add” and nearby direction words—that mark step-by-step guidance.

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

commodation

-1.87

complished

-1.84

⭏

-1.80

COLOG

-1.77

ༀ

-1.76

🤓

-1.73

ceptable

-1.66

ньому

-1.62

🤙

-1.61

🌞

-1.60

POSITIVE LOGITS

1.84



1.66

ーと

1.65

1.62

 precludes

1.59

跸

1.57

 minus

1.57

同樣

1.56

很想

1.55

當天

1.53

Activations Density 0.053%

adding configuration or elements

the word "the" in technical or instructional contexts where it precedes specific items to be added, modified, or referenced.

The neuron fires on the imperative “how‐to” instruction phrases—e.g. verbs like “download” or “add” and nearby direction words—that mark step-by-step guidance.

No Comments

No Known Activations

adding configuration or elements

the word "the" in technical or instructional contexts where it precedes specific items to be added, modified, or referenced.

The neuron fires on the imperative “how‐to” instruction phrases—e.g. verbs like “download” or “add” and nearby direction words—that mark step-by-step guidance.

No Comments

No Known Activations