RL新思路!复旦用游戏增强VLM通用推理,性能匹敌几何数据
3天前新智元报道 编辑:LRST 【新智元导读】复旦大学NLP实验室研发Game-RL,利用游戏丰富视觉元素和明确规则生成多模态可验证推理数据,通过强化训练提升视觉语言模型的推理能力。创新性地提出C …
停止RL研究!前OpenAI研究员:互联网才是唯一重要的技术
5个月前A former OpenAI researcher urges to halt reinforcement learning studies, emphasizing that the internet remains the ultimate technological force. This perspective sparks debate over the future priorities in AI development and digital innovation.