Zephaniah Roe

Karma: 904

Formally zroe1

Second Look Research / XLab

https://github.com/zroe1

Trees are mostly made of air and a generalizable lesson for AI safety

Zephaniah Roe29 May 2026 4:08 UTC

169 points

28 comments4 min readLW link

Iterative Finetuning is Mostly Idempotent

Zephaniah Roe, jcksanderson and Julian H

11 May 2026 6:41 UTC

23 points

0 comments5 min readLW link

Summer AI Safety Opportunities at UChicago XLab

Zephaniah Roe9 Mar 2026 6:26 UTC

27 points

0 comments3 min readLW link

Principles for Meta-Science and AI Safety Replications

Zephaniah Roe23 Jan 2026 6:59 UTC

47 points

7 comments4 min readLW link

What Washington Says About AGI

Zephaniah Roe17 Jan 2026 5:43 UTC

134 points

7 comments6 min readLW link

Introducing the XLab AI Security Guide

Zephaniah Roe, jcksanderson and Julian H

27 Dec 2025 16:50 UTC

19 points

1 comment5 min readLW link

Against “You can just do things”

Zephaniah Roe8 Nov 2025 0:58 UTC

61 points

9 comments3 min readLW link

zroe1′s Shortform

Zephaniah Roe20 Sep 2025 21:19 UTC

2 points

45 comments1 min readLW link

Intriguing Properties of gpt-oss Jailbreaks

Zephaniah Roe and jcksanderson

13 Aug 2025 19:42 UTC

19 points

0 comments10 min readLW link

(xlabaisecurity.com)

Alternative Models of Superposition

Zephaniah Roe and RGRGRG

11 Aug 2025 15:52 UTC

20 points

6 comments5 min readLW link