INDEX

Explanations

code tokens and identifiers

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

variable names, identifiers, or namespace paths in code that use underscores or dots as separators.

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

The neuron detects tokens that recur multiple times in a local snippet (i.e. repeated identifiers), with its activation growing each time the same token appears.

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

苼

-1.40

 ΤΟ

-1.34

を使って

-1.34

continues

-1.32

ſſen

-1.30

 těch

-1.29

🡺

-1.29

followed

-1.29

ficador

-1.27

 exemplu

-1.27

POSITIVE LOGITS

is

1.78

 would

1.59

are

1.47

 jutaan

1.47

 which

1.46

my

1.38

”，

1.36

The

1.36

had

1.30

 були

1.28

Activations Density 0.054%

code tokens and identifiers

variable names, identifiers, or namespace paths in code that use underscores or dots as separators.

The neuron detects tokens that recur multiple times in a local snippet (i.e. repeated identifiers), with its activation growing each time the same token appears.

No Comments

No Known Activations

code tokens and identifiers

variable names, identifiers, or namespace paths in code that use underscores or dots as separators.

The neuron detects tokens that recur multiple times in a local snippet (i.e. repeated identifiers), with its activation growing each time the same token appears.

No Comments

No Known Activations