INDEX

Explanations

threatening language and confrontational interactions

oai_token-act-pair · gpt-3.5-turbo

Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_fs768-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_768

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

768

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

IFrame

Link

Negative Logits

 broadly

-0.86

 markedly

-0.85

 strikingly

-0.83

xtap

-0.80

 outset

-0.77

 Historically

-0.76

 reliance

-0.75

 bolstered

-0.74

 policymakers

-0.74

 principally

-0.72

POSITIVE LOGITS

 fuckin

1.65

 fucking

1.52

 shit

1.50

 gonna

1.40

 bitch

1.39

 crap

1.35

 fucked

1.35

 fuck

1.32

 shitty

1.31

 haha

1.30

Activations Density 11.023%

threatening language and confrontational interactions

No Known Activations