INDEX

Explanations

client followed by code constructs

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

the word "Client" in programming code, particularly in class names, variable names, and import statements.

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

The neuron strongly activates on occurrences of the token “Client,” i.e. references to a Client class or identifier.

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

↵

-3.30

</strong>

-3.11

-2.86

“

-2.84

-2.52

“

-2.50

-2.42

-2.36

要不然

-2.30

趿

-2.23

POSITIVE LOGITS



3.03

इटम

2.95

渽

2.73

 этом

2.64

個月

2.56

 kerana

2.52

 này

2.50

 olvidado

2.47

沇

2.44

 This

2.42

Activations Density 0.018%

client followed by code constructs

the word "Client" in programming code, particularly in class names, variable names, and import statements.

The neuron strongly activates on occurrences of the token “Client,” i.e. references to a Client class or identifier.

No Comments

No Known Activations

client followed by code constructs

the word "Client" in programming code, particularly in class names, variable names, and import statements.

The neuron strongly activates on occurrences of the token “Client,” i.e. references to a Client class or identifier.

No Comments

No Known Activations