personal-agentIMAGE
EPFL 论文把 Grok Bot 当量化研究操作系统
论文提出模型生成想法、决策层判断置信度、确定性代码管风险,按置信度分流执行或人工复核。
kvro
@0xkvro27 天前收录原推发布于 2026.10.07

kvro @0xkvro原推发布于 2026.10.07
this EPFL paper is f*cking insane it treats Grok Bot like an operating system for quant research not a chatbot the stack is simple: model → generates ideas decision layer → judges confidence deterministic code → enforces risk then the system routes every action: high confidence → execute medium confidence → verify more low confidence → human review the key idea: the model explores the system decides that is how Grok Bot becomes infrastructure instead of another AI demo bookmark before this gets buried
这条是引用着另一条写的:x.com/0xkvro/status/2106763188100362388
x.com/0xkvro/status/2107509266848289005采集于 2026.10.07 10:27