INDEX

Explanations

words related to dishonesty and disgraceful behavior

oai_token-act-pair · gpt-3.5-turbo

terms related to dishonesty, disgrace, and betrayal

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

uana

-0.80

ļéĨĴ

-0.78

 liner

-0.73

anes

-0.72

 parcels

-0.71

oan

-0.70

opot

-0.69

 applic

-0.67

ynt

-0.67

uli

-0.66

POSITIVE LOGITS

 disgrace

1.13

 dishon

1.05

 unfocusedRange

0.92

 traitor

0.87

 honorable

0.86

Hon

0.81

brate

0.80

ously

0.76

 shameful

0.76

BILITY

0.75

Activations Density 0.024%

words related to dishonesty and disgraceful behavior

terms related to dishonesty, disgrace, and betrayal

No Comments

No Known Activations