INDEX

Explanations

trial and error

The neuron fires on words and phrases that denote experimentation or trial-and-error processes.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

-1.47

MAXN

-1.34

)：

-1.25

 damals

-1.23

「

-1.21

𝟏

-1.21

たちは

-1.20

 бъ

-1.19

 ursprünglich

-1.18

 бампер

-1.16

POSITIVE LOGITS

हे

1.38

 TIPO

1.38

ૅ

1.36

."

1.31

}$.

1.30

 INFORMACIÓN

1.30

 estrecha

1.27

.”

1.27

 while

1.25

)}$

1.23

Activations Density 0.049%