RSS

Puria

Karma: 512

I’m helping build geodesicresearch.ai

Personal writing at puriaradmard.substack.com

Why study proto-train­ing gam­ing as an ad­ver­sar­ial al­ign­ment failure mode?

Puria, Edward James Young and Cam
8 Jul 2026 17:07 UTC
55 points
0 comments7 min readLW link

Why study al­ign­ment in­ter­ven­tions on pre-RL check­points?

Edward James Young, Puria and Cam
8 Jul 2026 17:07 UTC
66 points
2 comments6 min readLW link

An­nounc­ing Geodesic Research

Puria, Cam, Alexandra Narin, Edward James Young and Kyle O’Brien
27 May 2026 16:40 UTC
85 points
2 comments5 min readLW link

Learned Chain-of-Thought Obfus­ca­tion Gen­er­al­ises to Unseen Tasks

Nathaniel Mitrani, sassanb, Cam and Puria
21 May 2026 10:11 UTC
31 points
0 comments5 min readLW link
(arxiv.org)

Align­ment Pre­train­ing: AI Dis­course Causes Self-Fulfilling (Mis)alignment

Cam, Puria, Kyle O’Brien, David Africa, Samuel Ratnam and andyk
21 Dec 2025 0:53 UTC
207 points
25 comments9 min readLW link

Ar­chi­tec­tures for In­creased Ex­ter­nal­i­sa­tion of Reasoning

Karthik Viswanathan, Liza Pavlova, Mariia Koroliuk, Puria, Cam and Edward James Young
26 Nov 2025 20:24 UTC
38 points
2 comments13 min readLW link

Gen­er­al­i­sa­tion Hack­ing: a first look at ad­ver­sar­ial gen­er­al­i­sa­tion failures in de­liber­a­tive alignment

Cam and Puria
17 Nov 2025 21:44 UTC
54 points
2 comments8 min readLW link

I Am Large, I Con­tain Mul­ti­tudes: Per­sona Trans­mis­sion via Con­tex­tual In­fer­ence in LLMs

Shi and Puria
8 Sep 2025 13:52 UTC
33 points
0 comments1 min readLW link
(www.researchgate.net)