INDEX

Explanations

terms related to misinformation or deceptive information

oai_token-act-pair · gpt-3.5-turbo

terms related to misleading or deceptive information

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.0.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

estones

-0.72

ahime

-0.72

erg

-0.72

emed

-0.71

ternal

-0.66

 neighbor

-0.64

 Beat

-0.63

github

-0.63

ero

-0.62

nder

-0.61

POSITIVE LOGITS

 misleading

3.53

 mislead

2.75

 deceptive

2.64

 misled

2.44

 dece

2.29

 deceive

2.05

 misrepresent

1.98

 deceived

1.91

 deception

1.90

 misinformation

1.73

Activations Density 0.034%

terms related to misinformation or deceptive information

terms related to misleading or deceptive information

No Comments

No Known Activations