INDEX

Explanations

reading and writing files

The neuron selectively fires on programming‐language tokens—symbols (like &, *, →), punctuation, and code identifiers (underscored names, data types, function names)—rather than on ordinary English words.

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

you

-0.98

 typical

-0.94

 with

-0.88

 visible

-0.88

 effects

-0.87

for

-0.85

Detected

-0.84

 Efficient

-0.84

 conspiracy

-0.82

↵

-0.82

POSITIVE LOGITS

 AstraZeneca

1.00

indelijk

0.96

同样的

0.96

automne

0.92

 Lamborghini

0.91

Зно

0.90

 चीज़ों

0.89

伫

0.88

 robar

0.88

splitContainer

0.87

Activations Density 0.008%