x

LESSWRONG

LW

Anish Tondwalkar — LessWrong

Anish Tondwalkar

Anish Tondwalkar

Message

Used to kick models in the head to see how they ticked at Google Brain and OpenAI. Now Co-Founder at dmodel.ai, a private mechinterp lab.

Outside of AI, my background is in formal methods and in high energy theory.

28

Ω

2

1

2

7y

Anish Tondwalkar

Used to kick models in the head to see how they ticked at Google Brain and OpenAI. Now Co-Founder at dmodel.ai, a private mechinterp lab.

Outside of AI, my background is in formal methods and in high energy theory.

Discovering Concept-Editing Algorithms With LLM Agents

by Adam Scherlis and Anish Tondwalkar

Concept erasure is a technique that removes unwanted information from a model’s activations, but current erasure methods struggle to fully remove target concepts. In this study, we tasked LLM agents trained on our data with inventing concept erasure algorithms that outperform current methods under the same experimental constraints. We measure...

How Language Models Understand Nullability

TL;DR Large language models have demonstrated an emergent ability to write code, but this ability requires an internal representation of program semantics that is little understood. Recent interpretability work has demonstrated that it is possible to extract internal representations of natural language concepts, raising the possibility that similar techniques could...

Mar 11, 2025•5