1
Alignment pretraining: AI discourse creates self-fulfilling (mis)alignment
研究发现AI discourse在预训练中会自我实现(mis)alignment,观点新颖有冲击力
Article URL: https://arxiv.org/abs/2601.10160 Comments URL: https://news.ycombinator.com/item?id=48185938 Points: 5 # Comments: 1