INDEX

Explanations

temporal clauses starting with when

The neuron strongly activates on very high‐frequency/common tokens (e.g. “and,” “as,” “glad”) and thus is detecting how often a token appears.

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

毎回

-1.33

haviours

-1.16

selbe

-1.13

 gedurende

-1.09

]^{

-1.05

\}=\

-1.04

inys

-1.04

顿

-1.03

iress

-1.02

とおり

-1.02

POSITIVE LOGITS

 when

12.56

 cuando

8.19

 когда

8.06

 quando

7.84

when

7.31

When

6.53

 όταν

6.25

 When

6.06

 عندما

6.03

 ketika

5.88

Activations Density 2.171%