INDEX

Explanations

caused by

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

code-related structures, particularly functions, programming syntax, and code elements across different languages.

oai_token-act-pair · claude-3-7-sonnet-20250219

Triggered by @neilrathi

The neuron fires on technical or domain-specific vocabulary (e.g. programming keywords, jargon, or specialized terms) across different languages.

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

Negative Logits

is

-1.63

极

-1.38

 homicidio

-1.34

 коротки

-1.33

 again

-1.31

อีก

-1.25

Quien

-1.23

If

-1.23

 Если

-1.22

Quién

-1.20

POSITIVE LOGITS

 these

1.77

 эти

1.54

 with

1.47

ală

1.45

<b>

1.38

these

1.34

 June

1.33

شویی

1.30

 czerwca

1.27

 gdyż

1.27

Activations Density 0.011%

caused by

code-related structures, particularly functions, programming syntax, and code elements across different languages.

The neuron fires on technical or domain-specific vocabulary (e.g. programming keywords, jargon, or specialized terms) across different languages.

No Known Activations

caused by

code-related structures, particularly functions, programming syntax, and code elements across different languages.

The neuron fires on technical or domain-specific vocabulary (e.g. programming keywords, jargon, or specialized terms) across different languages.

No Known Activations