INDEX

Explanations

toxic anterior segment syndrome

np_acts-logits-general · gemini-2.5-flash-lite

programming code structure and JSON-like content with braces and indentation patterns.

oai_token-act-pair · claude-3-7-sonnet-20250219 Triggered by @neilrathi

The neuron selectively detects indentation whitespace tokens (i.e. runs of spaces used to indent structured or nested text).

oai_token-act-pair · o4-mini Triggered by @jyhe0408

New Auto-Interp

Configuration

google/gemma-scope-27b-pt-res/layer_22/width_131k

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 several

-1.43

 many

-1.17

//!

-1.16

had

-1.16

 very

-1.16

 greatly

-1.14

 deeply

-1.13

 totally

-1.12

so

-1.12

you

-1.12

POSITIVE LOGITS

ICTS

1.35

Οι

1.31

اگر

1.30

 "]";

1.30

Après

1.25

 també

1.20

 RELIGION

1.20

 też

1.20

วันที่

1.20

ὴν

1.19

Activations Density 0.000%

toxic anterior segment syndrome

programming code structure and JSON-like content with braces and indentation patterns.

The neuron selectively detects indentation whitespace tokens (i.e. runs of spaces used to indent structured or nested text).

No Comments

No Known Activations

toxic anterior segment syndrome

programming code structure and JSON-like content with braces and indentation patterns.

The neuron selectively detects indentation whitespace tokens (i.e. runs of spaces used to indent structured or nested text).

No Comments

No Known Activations