x

LESSWRONG

LW

Advik

Advik

Message

I like breaking models.

Researching red-teaming as part of the MATS 10.0 program with UK AISI. Dabbled a bit with mechanistic interp, blue-teaming and reverse engineering; I'm especially interested in discovering vulnerabilities and flaws in generative models and the science of post-training.

https://www.a3v1k.com/ :)

19

1

1

7mo

Advik

I like breaking models.

Researching red-teaming as part of the MATS 10.0 program with UK AISI. Dabbled a bit with mechanistic interp, blue-teaming and reverse engineering; I'm especially interested in discovering vulnerabilities and flaws in generative models and the science of post-training.

https://www.a3v1k.com/ :)

Advik — LessWrong

Brittle model organisms obstructs deception elicitation work

This work was done by Advik Raj Basani with Daniel Tan and Chloe Li as part of SPAR Spring 2026. tl;dr: Finetuning-based auditing methods for model organisms may unintentionally erase the deceptive behavior we are trying to measure, leading to an illusion of success. We study secret side constraint (SSC)...