Recent News
05/2024
We have made significant updates to the English version of the paper (V5)!
- Added outstanding papers related to alignment at ICLR 2024, covering both spotlight and oral sessions.
- Enhanced the focus on major contradictions, core scientific challenges, and hot topics by analyzing the Key Challenges in the Alignment Cycle, including learning human intent from rich modalities, trustworthy tools for assurance, and the processes of value elicitation and implementation.
- Summarized the key traits and future directions in alignment research, emphasizing policy relevance among other aspects.
- Added numerous articles and tracks related to the recent MATS program.
02/2024
We have made significant updates to the English version of the paper (V4)!
- Added significant work since November, such as the latest work from ICLR 2024.
- Added content related to the Singapore Conference on AI (SCAI).
- Completed a meticulous filtering of the references, updated the citation format for papers accepted by top conferences, and conducted a strict selection of the quality of the references, ultimately determining 801 references.
- Visualized the years of the references to show the development trend in the alignment. Details can be referred to here.
- Added discussions on Weak-to-Strong Generalization (Section 2.5), specifically outlining its basic process, key concepts, various ideas and technical solutions combining it with Scalable Oversight.
- Added discussions on AI alignment from a socio-technical perspective (Section 5.4).
Alignment Survey has been recognized by NIST!
- The National Institute of Standards and Technology (NIST), under the U.S. Department of Commerce, has adopted the Alignment Cycle framework proposed by the Alignment Survey in its Trusted and Responsible AI research program. For more details, refer to here.
Alignment Survey has received coverage and attention from multiple media outlets!
- Alignment Survey was featured as a cover story by Sanlian Life Weekly(三联生活周刊).
- Alignment Survey was reported by Banyuetan(半月谈) as a representative work in AI alignment.
- Alignment Survey was successively reported by several AI media, including Synced(机器之心), QbitAI(量子位), and Concordia AI(安远AI).