INDEX

Explanations

German government bonds

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

The neuron is essentially a “position‐in‐the‐text” signal: it fires most strongly on tokens near the start and then its activation decays as you move further into the document.

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

words or text fragments that are part of technical documentation, code, product listings, or structured data formats.

oai_token-act-pair · claude-4-5-sonnet Triggered by @jyhe0408

markers that denote lists, ranges, or options, such as punctuation and conjunctions used to separate items or indicate spans

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

New Auto-Interp

Configuration

google/gemma-scope-2-12b-pt/resid_post/layer_24_width_16k_l0_medium

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ijų

0.83

ye

0.83

𝘨

0.82

iezan

0.81

pti

0.80

mia

0.77

%")

0.77

hiyam

0.77

kti

0.77

<0x8D>

0.76

POSITIVE LOGITS

0.86

↵

0.74

 slowdown

0.73

צא

0.73

(<

0.71

 Puede

0.71

 goodwill

0.71

 Pode

0.70

use

0.70

אם

0.68

Activations Density 0.000%

German government bonds

The neuron is essentially a “position‐in‐the‐text” signal: it fires most strongly on tokens near the start and then its activation decays as you move further into the document.

words or text fragments that are part of technical documentation, code, product listings, or structured data formats.

markers that denote lists, ranges, or options, such as punctuation and conjunctions used to separate items or indicate spans

No Comments

No Known Activations