INDEX

Explanations

statement result conclusion

The neuron fires on words that introduce or label the outcome of a proof or argument—e.g. “claim,” “conclusion,” “result,” “desired.”

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

🤴

-1.37

幸好

-1.31

slightly

-1.31

tersebut

-1.30

 prêtre

-1.28

idlertid

-1.27

 ainsi

-1.27

⩥

-1.27

unately

-1.26

嗩

-1.23

POSITIVE LOGITS

 from

1.84

 more

1.60

all

1.53

 with

1.49

。

1.46

 both

1.45

。“

1.41

any

1.40

 only

1.31

 when

1.30

Activations Density 0.009%