INDEX

Explanations

end of jacobian

The neuron strongly activates on numeric tokens (years, measurements, citation numbers, etc.), i.e. it detects numbers.

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

 would

1.02

would

0.93

 Would

0.91

Would

0.89

 WOULD

0.84

 wouldn

0.80

 serait

0.78

 sarebbe

0.75

 seharusnya

0.75

 oleks

0.75

POSITIVE LOGITS

))

2.80

()))

2.72

 ""))

2.53

"))

2.50

));

2.47

'))

2.47

()));

2.45

))

2.31

."))

2.28

)),

2.27

Activations Density 0.030%