INDEX

Explanations

phrases related to a controversial or offensive event and the subsequent actions taken

oai_token-act-pair · gpt-3.5-turbo

instances of apologies and acknowledgment of mistakes

oai_token-act-pair · gpt-4o-mini

Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 10-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.10.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.10.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Negative Logits

cells

-0.88

 trillions

-0.83

omics

-0.82

 cures

-0.81

arbon

-0.81

 fert

-0.76

 Recovery

-0.75

 forecasts

-0.75

 forecasting

-0.74

 savings

-0.73

POSITIVE LOGITS

 offended

1.50

 offending

1.33

 homophobic

1.33

 objectionable

1.32

 Yiannopoulos

1.28

 boycott

1.26

 disrespectful

1.23

 sexist

1.21

 slurs

1.20

 discriminatory

1.18

Activations Density 1.327%

phrases related to a controversial or offensive event and the subsequent actions taken

instances of apologies and acknowledgment of mistakes

No Known Activations