INDEX

Explanations

instances of the word "mistaken" and its variations, frequently in contexts implying confusion or erroneous beliefs

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_fs24576-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_24576

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.8.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 horm

-0.75

ighth

-0.74

inth

-0.72

raq

-0.71

onse

-0.69

ajo

-0.67

ŃĶ

-0.67

essen

-0.66

rollers

-0.65

atever

-0.64

POSITIVE LOGITS

ously

0.75

 mistaken

0.69

 Brand

0.67

 Desire

0.66

ame

0.62

 inexper

0.62

 Identity

0.62

MIS

0.62

opsis

0.61

cair

0.61

Activations Density 0.026%

instances of the word "mistaken" and its variations, frequently in contexts implying confusion or erroneous beliefs

No Comments

No Known Activations