INDEX

Explanations

phrases indicating negative or concerning news contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L5_Hcat_z_lr1.20e-03_l11.00e+00_ds49152_bs4096_dc1.00e-06_rsanthropic_rie25000_nr4_v9.pt

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

49,152

Data Type

float32

Hook Name

blocks.5.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Head Attr Weights

0:0.02

1:0.01

2:0.20

3:0.10

4:0.12

5:0.02

6:0.02

7:0.25

8:0.03

9:0.04

10:0.08

11:0.07

Negative Logits

displayText

-1.77

ioned

-1.73

ationally

-1.71

acebook

-1.69

vernment

-1.67

externalActionCode

-1.60

omial

-1.56

soType

-1.54

iology

-1.49

reason

-1.49

POSITIVE LOGITS

 overhe

1.58

 politic

1.57

 anecdotal

1.51

 wedge

1.50

 anecd

1.48

 piling

1.43

 woes

1.42

 dred

1.38

 newfound

1.33

getic

1.33

Activations Density 0.001%

phrases indicating negative or concerning news contexts

No Comments

No Known Activations