What are chinese labs doing differently?
Chinese AI models are reportedly improving rapidly despite lower spending compared to American labs. One suggested explanation is their acquisition of handcrafted expert data from US companies such as SurgeAI and Mercor. Some believe that access to such data should be restricted, similar to chips and GPUs, to help the US maintain its lead in the AI race.
时间与来源
时间显示为 UTC
显示时区：UTC
本地时区尚不可用，暂时显示 UTC。
发布当时偏移：UTC+02026年9月29日 19:55 UTC
收录当时偏移：UTC+02026年9月30日 01:00 UTC
- 发布
- 2026年9月29日 19:55
- 收录
- 2026年9月30日 01:00
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
档位是按信源手工设定的编辑判断，不是逐条打分。
Chinese models seem to keep getting better while only spending a fraction of what American labs do and i’m curious what the actual explanation is.
Is it better efficiency? Better post-training? Better use of open research? Better data?
I know recently they’ve been buying up tons of handcrafted expert data from US companies like SurgeAI/Mercor. I think stuff like this should be restricted, like chips/GPUs, if we want the US to maintain the lead in the AI race.
What things could it be?