INDEX

Explanations

instances of the word "secretly" and similar phrases indicating covert actions

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res_fs49152-jb

Configuration

jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_49152

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

49,152

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 GOODMAN

-0.90

nesota

-0.86

Pwr

-0.82

ERAL

-0.77

orem

-0.76

EVA

-0.76

 Citation

-0.74

ĸļ

-0.72

course

-0.71

ces

-0.70

POSITIVE LOGITS

 tucked

0.83

 taped

0.83

 buried

0.83

 plotted

0.80

 transmitted

0.78

 guarded

0.74

 resent

0.74

 sabot

0.74

 housed

0.73

 suppressed

0.73

Activations Density 0.016%

instances of the word "secretly" and similar phrases indicating covert actions

No Comments

No Known Activations