INDEX

Explanations

phrases prompting for input

The neuron fires on occurrences of the prompt “Enter” (and its immediate context) in user‐input or I/O prompt strings.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 печени

-1.60

 этих

-1.57

isn

-1.46

 attirer

-1.45

 complicado

-1.43

 perfectos

-1.41

 áng

-1.38

または

-1.35

 помогает

-1.34

inicial

-1.33

POSITIVE LOGITS

if

1.63

any

1.59

 during

1.52

 after

1.48

 Both

1.47

 what

1.41

</td>

1.41

 some

1.36

 when

1.34

 both

1.33

Activations Density 0.009%