x

LESSWRONG

LW

Jasper Chong — LessWrong

Jasper Chong

Jasper Chong

Message

I study mathematics and computer science, and am interested in how AI works.

I believe that understanding how these models work mathematically, will give rise to new and powerful safety methods.

Here is my website! Would love to hear from you: https://www.jasperchong.com/

19

1

4mo

Jasper Chong

I study mathematics and computer science, and am interested in how AI works.

I believe that understanding how these models work mathematically, will give rise to new and powerful safety methods.

Here is my website! Would love to hear from you: https://www.jasperchong.com/

SONI: Selective Orthogonalisation via Noise Injection

This project was completed as a capstone for TARA. All code is available in github. TL;DR * The Problem: Neural networks use superposition to pack many concepts into small latent spaces by making feature vectors almost-orthogonal. This entanglement makes models opaque and breaks safety interventions (e.g. concept erasure, activation steering)...