Competing with sampling — Alignment Research Center

In 2025, ARC has been making conceptual and theoretical progress at the fastest pace that I've seen since I first interned in 2022. Most of this progress has come about because of a re-orientation around a more specific goal: outperforming random sampling when it comes to understanding neural

In 2025, ARC has been making conceptual and theoretical progress at the fastest pace that I've seen since I first interned in 2022. Most of this progress has come about because of a re-orientation around a more specific goal: outperforming random sampling when it comes to understanding neural network outputs. Compared to our previous goals, this goal has the advantage of being more concrete and more directly tied to useful applications. The purpose of this post is to: Explain and motivate our "outperforming sampling" agenda from the standpoint of preventing catastrophic AI misalignment. Introd

saved by

• Gloria Ma
• Yudhister Joel Kumar
• Qiao Zhang
• Sam B
• Anonymous Bat

related reading

• ARC progress update: Competing with sampling — LessWronglesswrong.com
• A Mike's-Eye View of ARC's Research — Alignment Research Centeralignment.org
• Tips for Empirical Alignment Research — AI Alignment Forumalignmentforum.org
• Alignment is not solved but it increasingly looks solvablealigned.substack.com
• Alignment Research Centeralignment.org
• Recommendations for Technical AI Safety Research Directionsalignment.anthropic.com
• The Universe from an Intentional Stancecasparoesterheld.com
• Can we safely automate alignment research? - Joe Carlsmithjoecarlsmith.com
• Formal verification, heuristic explanations and surprise accounting — Alignment Research Centeralignment.org
• An Ambitious Vision for Interpretability — AI Alignment Forumalignmentforum.org
• A Pragmatic Vision for Interpretability — AI Alignment Forumalignmentforum.org
• [2605.05179] Estimating the expected output of wide random MLPs more efficiently than samplingarxiv.org