Welcome to Alignment Survey Materials 🎉!
Why Alignment Survey Materials?
During the alignment survey process, we gained insights into various alignment techniques and the phenomenon of misalignment. This material will summarize the key points of the paper, presenting our research findings in a more accessible and reader-friendly manner.
We have curated a number of alignment-related papers (continuously updated) and indexed them using a user-friendly tree diagram, enabling you to explore specific areas within alignment quickly!
What can I Find?
- Quick Start: A brief introduction to the motivation of alignment and the concept, methodology, and practice of AI alignment, highlighting its potential future directions.
- Learning from Feedback: A collection of papers and blogs related to our proposed Learning from feedback, which concerns the question of during forward alignment, how do we provide and use feedback to any given behavior (of the trained AI system) under any given input?.
- Learning under Distribution Shift: Our discourse on generalization by incorporating an exhaustive exploration of distributional shift, especially its intrinsic relationship with AI alignment.
- Assurance: An introduction to the measurement and evaluation of the AI system’s practical alignment after the AI system is trained and deployed.
- Governance: A survey and compilation of research on the role, historical studies, and open problems in AI governance.
- FAQ: A collection of frequently asked questions about alignment, enabling readers to acquire fragmented knowledge about alignment quickly.