注册并分享邀请链接,可获得视频播放与邀请奖励。

Brett Caughran
@FundamentEdge
Training Humans at Fundamental Edge Training Agents at Fundamental Research Labs
加入 September 2018
4.6K 正在关注    63.3K 粉丝
Hard to know for sure, but this feels right to me. The alpha isn't in post-training the model; it's in the execution harness and skill layer orchestrating it. Hard part is the proprietary Evals and modularizing the Skill architecture to give the system incisive judgment and make it feel intuitive to use.
显示更多
I unfortunately think superforecasting teams who are post training open weight models are going to get modelmogged by frontier labs that are obviously also working on forecasting benchmarks.