<![CDATA[皮哈皮阿]]> https://blog.phpat.com/ zh-cn 动态拼批不是开个开关：线上拼批的取舍与门禁 https://blog.phpat.com/ai-engineering/622.html 阅读全文]]> Sun, 25 Oct 2026 20:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-engineering/622.html Agent 子目标检查点回归：从回滚状态到登记副作用 https://blog.phpat.com/agent-engineering/621.html 阅读全文]]> Sun, 25 Oct 2026 14:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/agent-engineering/621.html 推理模型输出被截断后怎么接回去 https://blog.phpat.com/ai-dev/620.html 阅读全文]]> Sun, 25 Oct 2026 09:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-dev/620.html 前缀缓存命中率：字节对齐与工程取舍 https://blog.phpat.com/ai-frontier/619.html 阅读全文]]> Sat, 24 Oct 2026 20:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-frontier/619.html 推理模型的流式输出：thinking 块怎么流 https://blog.phpat.com/ai-engineering/618.html 阅读全文]]> Sat, 24 Oct 2026 14:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-engineering/618.html Agent 轨迹回放调试：把失败变成可回放资产 https://blog.phpat.com/agent-engineering/617.html 阅读全文]]> Sat, 24 Oct 2026 09:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/agent-engineering/617.html Agent 上下文压缩：四层策略与失败代价 https://blog.phpat.com/ai-dev/616.html 阅读全文]]> Fri, 23 Oct 2026 20:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-dev/616.html 投机解码落地：草稿模型对齐与逐利陷阱 https://blog.phpat.com/ai-frontier/615.html 阅读全文]]> Fri, 23 Oct 2026 14:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-frontier/615.html 长度惩罚参数调优：常见参数名字容易混淆 https://blog.phpat.com/ai-engineering/614.html 阅读全文]]> Fri, 23 Oct 2026 09:01:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-engineering/614.html 并行工具调用暴露的工具设计耦合问题 https://blog.phpat.com/agent-engineering/613.html 阅读全文]]> Fri, 23 Oct 2026 09:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/agent-engineering/613.html 约束解码让 JSON 不再崩，但它会改坏推理 https://blog.phpat.com/ai-dev/612.html 阅读全文]]> Thu, 22 Oct 2026 20:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-dev/612.html 思维链压缩到 20%，还能保住数学推理的准确率吗 https://blog.phpat.com/ai-frontier/611.html 阅读全文]]> Thu, 22 Oct 2026 14:15:27 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-frontier/611.html 世界模型被吹了两年，真正难的不是生成视频 https://blog.phpat.com/ai-frontier/world-models-beyond-video.html 阅读全文]]> Tue, 20 Oct 2026 20:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-frontier/world-models-beyond-video.html AI 浏览器用了一个月，我还是回到了地址栏 https://blog.phpat.com/ai-frontier/ai-browser-address-bar.html 阅读全文]]> Tue, 20 Oct 2026 19:20:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-frontier/ai-browser-address-bar.html 个人 Agent 与常驻智能体：长期记忆、主动性与隐私边界 https://blog.phpat.com/ai-frontier/personal-agents-memory-privacy.html 阅读全文]]> Tue, 20 Oct 2026 10:30:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-frontier/personal-agents-memory-privacy.html 2026 年 AI 应用评测：从单轮打分到多轮轨迹评估 https://blog.phpat.com/ai-engineering/ai-evaluation-trajectory.html 阅读全文]]> Mon, 19 Oct 2026 09:40:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-engineering/ai-evaluation-trajectory.html 多智能体不是银弹：编排、通信与成本取舍 https://blog.phpat.com/ai-frontier/multi-agent-tradeoffs.html 阅读全文]]> Sun, 18 Oct 2026 10:00:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-frontier/multi-agent-tradeoffs.html Agent 安全清单：提示注入、工具越权与数据外泄 https://blog.phpat.com/agent-engineering/agent-security-checklist.html 阅读全文]]> Fri, 16 Oct 2026 09:30:00 +0800 戴墨镜的刘备 https://blog.phpat.com/agent-engineering/agent-security-checklist.html AI Coding Agent 如何稳定提交可合并的 PR https://blog.phpat.com/agent-engineering/ai-coding-agent-mergeable-pr.html 阅读全文]]> Wed, 14 Oct 2026 10:10:00 +0800 戴墨镜的刘备 https://blog.phpat.com/agent-engineering/ai-coding-agent-mergeable-pr.html RAG 的下一站：Agentic Retrieval https://blog.phpat.com/ai-dev/agentic-retrieval.html 阅读全文]]> Mon, 12 Oct 2026 09:50:00 +0800 戴墨镜的刘备 https://blog.phpat.com/ai-dev/agentic-retrieval.html