INDEX

Explanations

and then a conclusion

The neuron is primarily detecting four‐digit numeric tokens (e.g. years or large numbers).

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

avía

-1.30

 EconPapers

-1.27

andı

-1.24

)!=

-1.24

aunque

-1.20

"/",

-1.18

.'</

-1.16

っていますが

-1.16

vertra

-1.16

clud

-1.15

POSITIVE LOGITS

 therefore

8.31

 thus

7.44

 hence

6.63

 поэтому

6.25

Therefore

5.47

因此

5.34

 Therefore

5.28

 quindi

5.22

therefore

5.06

 donc

4.97

Activations Density 0.167%