INDEX

Explanations

cannot fulfill requests

The neuron fires on FAQ‐style lead‐ins, especially the phrase “answers to [questions/queries]” marking Q&A sections.

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

0.94

sher

0.84

 elett

0.82

delicious

0.82

xk

0.80

ление

0.78

0.77

xz

0.77

 carriera

0.76

esor

0.76

POSITIVE LOGITS

 questions

1.25

 QUESTIONS

1.10

 perguntas

1.09

 queries

1.05

 सवालों

0.93

questions

0.91

 requests

0.91

 Questions

0.89

 سوال

0.89

 주제

0.88

Activations Density 0.034%