INDEX

Explanations

computer code snippets

The neuron strongly activates on the recurring “download the pictures in high quality” instruction, i.e. it detects download‐image directives.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 CHAP

0.83

が多く

0.77

 admissible

0.75

 Liabilities

0.74

對

0.73

 "***

0.73

 structures

0.72

ৈতিক

0.72

ორცი

0.72

 sở

0.71

POSITIVE LOGITS

this

1.13

all

1.03

you

0.97

when

0.89

replaced

0.86

also

0.85

there

0.80

while

0.80

we

0.79

act

0.79

Activations Density 0.004%