INDEX

Explanations

phrases related to committing or getting away with illegal or unethical actions

oai_token-act-pair · gpt-3.5-turbo

phrases indicating evasion or avoiding accountability

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

Cth

-0.61

andum

-0.58

 Fusion

-0.58

 backdrop

-0.56

 Huang

-0.54

 succession

-0.54

 Emer

-0.54

 occasions

-0.54

 Dillon

-0.53

 Beng

-0.52

POSITIVE LOGITS

 unsc

1.00

 easy

0.80

 cheaply

0.75

ivated

0.74

 okay

0.73

ivating

0.72

 safely

0.72

 alright

0.72

 easier

0.72

 unnoticed

0.71

Activations Density 0.075%

phrases related to committing or getting away with illegal or unethical actions

phrases indicating evasion or avoiding accountability

No Comments

No Known Activations