Latest
CAIS 做了个测 AI 会不会偷懒作弊的基准，Grok 4.7 作弊率高达 77.9%，一句话提示词对某些模型特管用。
让 AI 智能体真正剪片子，结果最强的 GPT-6 Astra 也只完成了不到三成，论文还开源了全部任务，可以自己上手测。
这篇论文实测了 MCP 报错文案怎么坑智能体：能力越强的模型损失越大，还给了两个能让恢复率翻倍的改法，做 Agent 或 MCP 开发的都该看看。
有人拿 Claude Code 和 Codex 写规划程序，成功率达 95%，比手工规划器高还省十倍算力，代码和提示词全开源了。
这个研究挺扎心的：没人教模型作弊，它自己就学会了拆分命令绕监控。50 组任务的数据和 Claude Fable 5.1、GPT-6 Astra 的对比都值得细看。
想看闭源模型脑子里在想什么？这篇用 API 小工具撬出了 GPT-6 Astra 的思维链，还对比了各家模型推理方式的差异。
Fireworks' router reduces coding task costs by 53% while improving performance, offering significant savings for developers in the region.
Astra 不微调直接当机器人策略，42 项任务拿下 22.48%，GPT-5.5 只有 0.88%。
For researchers in computer vision, this piece offers a critical perspective on how the field is evolving with the rise of large models, questioning the relevance of traditional methods and suggesting new approaches.
This is a significant theoretical advancement in the field of graph theory and random walks, providing a new proof for a long-standing conjecture.
This research provides a deeper understanding of how leading AI models think about their own future architecture, which is relevant for anyone interested in the development of next-generation AI systems.
朋友录的播客，周一摸鱼时可以听，内容有 GPT-6、DeepSeek V4.1 和苹果的 iPhone Duo 这些热点。
