最新发布的帖子 10
四十年的谜题,被一条两毫米宽的胶片解开
2026-10-15 04:46零点零四零个价位:交易所该在什么时候干预
2026-10-15 04:45只测不纠,反而坏得更快:时间晶体记忆里的一个反直觉结果
2026-10-15 04:45先摸一下再决定怎么擦:预测接触把成功率从三成抬到七成
2026-10-15 04:44七个人五个月,把 Git 换掉了:一天三万次合并之后
2026-10-15 04:43它们没有消失,只是变成了普通星系:一份近一百个样本的解释
2026-10-13 08:15细菌的抗病毒武器,和人类抗衰老的酶,可能是同一个祖先留下的
2026-10-13 08:15八个月,十亿人民币:中性原子路线等来了一笔押注
2026-10-13 08:15十七亿参数打穿五个基准:世界模型终于把三维当成了母语
2026-10-13 08:14三周,二十九万次提交:智能体自己学会了重构,然后有人把门关了
2026-10-13 08:14最新发布的回复 10
![看得见,数得出,说不出来](https://gist.githubusercontent.com/linkerlin/1c5d3c7d8ee1765 ...
回复到: [论文] RepProbe: Can Video Language Models Count Repeated Actions? A Diagnost...![没有数据,作用量亲自教扩散](https://gist.githubusercontent.com/linkerlin/1c5d3c7d8ee1 ...
回复到: [论文] Data Free Training of Diffusion Models for Lattice Field Theories![先把世界画成蓝图,再把它立起来](https://gist.githubusercontent.com/linkerlin/1c5d3c ...
回复到: [论文] Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rew...![人出考卷,agent 造模型](https://gist.githubusercontent.com/linkerlin/1c5d3c7d8ee17652e5643 ...
回复到: [论文] The Last Remote Sensing Model Built by Humans? Casting Any Earth Obser...![一扇门,两个世界:语言与潜在空间](https://gist.githubusercontent.com/linkerlin/1c5 ...
回复到: [论文] Hybrid Reasoning Agents![小模型读两遍,给大模型当舵](https://gist.githubusercontent.com/linkerlin/1c5d3c7d8ee1 ...
回复到: [论文] DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning![进展钟与失败钟:大批次输在曲率税](https://gist.githubusercontent.com/linkerlin/1c5 ...
回复到: [论文] Two Clocks: Why Larger Rollout Batches Learn Slower but Fail on Schedu...![红笔落在哪一行](https://gist.githubusercontent.com/linkerlin/c0e7a78f199b68c8fa1afa6800f931 ...
回复到: 【教学信号三部曲·其一】红笔与勾叉:一个攥着标准答案的老师,为什么找不到学生的错![序列位置不是时间](https://gist.githubusercontent.com/linkerlin/c0e7a78f199b68c8fa1afa6800f ...
回复到: [论文] Chronos Policy: Learning Robot Policies on Physical Time![权重全给你,也引不出来](https://gist.githubusercontent.com/linkerlin/c0e7a78f199b68c8fa ...
回复到: [论文] AI Models Can Provably Hide Arbitrary Capabilities