Blog

2026

• An analysis of AI-generated content at the Mechanistic Interpretability Workshop

  Andy Arditi, Iván Arcuschin, Andrew Lee

  July 2026ai

• Thoughts on Claude Fable's silent safeguards

  Andy Arditi

  June 2026 · LessWrongai

• In-context learning of representations can be explained by induction circuits

  Andy Arditi

  March 2026 · ICLR Blogpostsinterpretability

2025

• Finding "misaligned persona" features in open-weight models

  Andy Arditi, Runjin Chen

  September 2025 · LessWronginterpretability

• Follow-up experiments on preventative steering

  Runjin Chen, Andy Arditi

  September 2025 · LessWronginterpretability

• Do models say what they learn?

  Andy Arditi, Marvin Li, Joe Benton, Miles Turpin

  March 2025 · LessWrongai

• Finding features causally upstream of refusal

  Daniel Lee, Eric Breck, Andy Arditi

  January 2025 · LessWronginterpretability

2024

• AI as systems, not just models

  Andy Arditi

  December 2024ai

• Unlearning via RMU is mostly shallow

  Andy Arditi, Bilal Chughtai

  July 2024 · LessWronginterpretability

2023

• Refusal mechanisms: initial experiments with Llama-2-7b-chat

  Andy Arditi, Oscar Obeso

  December 2023 · LessWronginterpretability

2022

• The anatomy of proof generation

  Andy Arditi, Ye Zhang

  December 2022crypto

• KZG in practice: polynomial commitment schemes and their usage in scaling Ethereum

  Andy Arditi

  October 2022 · Scrollcrypto

• Zero-knowledge: theoretical foundations II

  Andy Arditi

  August 2022crypto

• Zero-knowledge: theoretical foundations I

  Andy Arditi

  July 2022crypto

• Stablecoins

  Andy Arditi

  May 2022crypto