INDEX

Explanations

but not

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

This neuron identifies numerical values that are contrasted with “but not” statements, i.e. numbers appearing in negated comparisons.

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

expressions of contrastive exclusion—statements that assert an effect or property and then specify an exception.

oai_token-act-pair · gpt-5 Triggered by @jyhe0408

The neuron activates strongly on the word "not" when it appears after "but" in scientific or technical text, particularly in contexts expressing contrast or exception (e.g., "but not").

oai_token-act-pair · claude-4-5-sonnet 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

-1.58

 entreg

-1.36

 juvenil

-1.34

仕方

-1.27

 eğer

-1.26

圣节

-1.25

 baloncesto

-1.23

 sterling

-1.23

察觉

-1.23

 víctima

-1.22

POSITIVE LOGITS

1.61

from

1.55

 others

1.54

droje

1.50

 برای

1.48

 that

1.48

 Здесь

1.44

řes

1.44

 conjunction

1.42

 other

1.37

Activations Density 0.026%

but not

This neuron identifies numerical values that are contrasted with “but not” statements, i.e. numbers appearing in negated comparisons.

expressions of contrastive exclusion—statements that assert an effect or property and then specify an exception.

The neuron activates strongly on the word "not" when it appears after "but" in scientific or technical text, particularly in contexts expressing contrast or exception (e.g., "but not").

No Comments

No Known Activations