RSS

Ziqian Zhong

Karma: 197

I do technical AI interp and safety research. https://​​fjzzq2002.github.io/​​

De­fault con­tinu­a­tion mes­sage in In­spect and Petri could be problematic

Ziqian Zhong11 Sep 2026 0:20 UTC
11 points
2 comments4 min readLW link

User aware­ness in fron­tier models

Ziqian Zhong and jsteinhardt
6 Aug 2026 20:43 UTC
59 points
1 comment12 min readLW link
(transluce.org)

Model self-iden­ti­fi­ca­tion could be sub­limi­nally transferred

Ziqian Zhong30 Jul 2026 9:47 UTC
27 points
0 comments6 min readLW link

Spon­ta­neous in­tro­spec­tion in out­put tampering

Ziqian Zhong26 Apr 2026 20:05 UTC
25 points
1 comment12 min readLW link

Pando: A Con­trol­led Bench­mark for In­ter­pretabil­ity Methods

Ziqian Zhong21 Apr 2026 21:40 UTC
6 points
0 comments3 min readLW link
(arxiv.org)

Ho­doscope: Vi­su­al­iza­tion for Effi­cient Hu­man Supervision

Ziqian Zhong and Shashwat Saxena
20 Feb 2026 23:41 UTC
9 points
0 comments2 min readLW link
(hodoscope.dev)

Im­pos­si­bleBench: Mea­sur­ing Re­ward Hack­ing in LLM Cod­ing Agents

Ziqian Zhong30 Oct 2025 2:52 UTC
63 points
6 comments3 min readLW link
(arxiv.org)

Weight-diff SVD for LLM Monitoring

Ziqian Zhong5 Aug 2025 0:31 UTC
2 points
0 comments2 min readLW link
(arxiv.org)