INDEX

Explanations

while loop constructs

The neuron is detecting loop‐and‐data‐fetch constructs in code (e.g. “while (…)”, “read”, “fetch_assoc”, etc.).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

and

-1.88

↵

-1.68

 because

-1.67

-1.64

is

-1.41

-1.40

’

-1.39

 doesn

-1.35

–

-1.34

-1.33

POSITIVE LOGITS

 miesto

1.80

 mieste

1.66

 biela

1.59

ｶ

1.55

купить

1.52

 verbess

1.52

荭

1.48

ᐕ

1.48

抱歉

1.48

 eenen

1.48

Activations Density 0.012%