INDEX

Explanations

end of code or text

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

closing braces and brackets in structured data formats like JSON and code.

oai_token-act-pair · claude-4-5-sonnet Triggered by @bfgcn7frf5

The neuron specializes in detecting numeric literals (e.g. standalone numbers in code or data).

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

New Auto-Interp

Configuration

google/gemma-scope-27b-pt-res/layer_10/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

 counterfeit

-1.30

 pudieron

-1.29

鹋

-1.25

ኹ

-1.23

vantaged

-1.23

 tenían

-1.20

 standout

-1.18

ngoing

-1.17

 lograron

-1.17

覚えて

-1.16

POSITIVE LOGITS

 augusti

1.41

 Página

1.27

on

1.23

)」

1.22

》,

1.20

餍

1.16

」,

1.14

it

1.13

🏥

1.09

 narration

1.07

Activations Density 0.148%

end of code or text

closing braces and brackets in structured data formats like JSON and code.

The neuron specializes in detecting numeric literals (e.g. standalone numbers in code or data).

No Comments

No Known Activations

end of code or text

closing braces and brackets in structured data formats like JSON and code.

The neuron specializes in detecting numeric literals (e.g. standalone numbers in code or data).

No Comments

No Known Activations