Modern AI models are aligned through post-training to adapt them to downstream tasks. Recent work shows that fine-tuning language models on narrow tasks can …
机构:苏黎世联邦理工
来源:arXiv 2609.35291 | AI4Papers 论文推荐平台