为什么同一段对话，用对 chat template 的模型表现天差地别？对话模型的一切行为都由一张「模板」定义：角色标记、特殊 token、系统提示的位置。本篇拆解 chat template 的结构、各家模板的差异、它如何决定「模型以为自己在跟谁说话」，以及微调与应用时最容易踩的模板坑。

Large language models (LLMs) are increasingly explored for graph understanding and decision-making, while System One mod…

Sep 26, 2026 Nano Banana Pro prompted by THE DECODER A new Nvidia paper describes a system that automatically optimizes …

Manuel Uth Sep 24, 2026 Anthropic's AI model Claude found a previously unknown enzyme system in DNA databases, doing mos…

In this **tutorial**, we work with **Jev**, TypeSafe AI’s first System One model, which does not generate text at all: w…

TypeSafe AI released Jev, a System One model that answers typed questions with probabilities instead of generating text.…

!(https://www.gstatic.com/images/branding/googleg/1x/googleg_standard_color_128dp.png)Add as a preferredsource on Google…

Vision-language-action (VLA) models face a timing gap between low-rate inference and high-rate robot execution. We chara…

Sep 29, 2026 OpenAI Key Points OpenAI is opening its internal plugin platform to outside developers and adding team feat…

At a glance Quine (opens in new tab) is a research effort to create a multimodal world model of biology and an interacti…

Contrastive-LM has released CLM-8B, the first open model in a new class called Contrastive Language Models (CLMs). CLM d…

Agentic memory is becoming essential for long-horizon AI agents, yet many existing systems rely on autoregressive LLMs t…

Natural interaction in digital and physical environments requires continuous perception and timely responses. Spoken dia…

Modern AI agents routinely cross trust boundaries: they ingest untrusted content, combine it with privileged instruction…

Sponsored Presented byVanguard A scalable AI system requires a strong foundation. Data, engineering, platforms, operatin…

Complex camera motion is central to visual storytelling, yet translating cinematographic intent into controllable, execu…

Recursive self-improvement (RSI) describes a system that improves not only its behavior but also the process by which it…

Improving every agent in an LLM-based multi-agent system can be prohibitively costly, making it important to identify wh…

Volvo is leveling up its automated driving system through an expansion of its nearly decade-long partnership with Nvidia…

At a glance StudentSim is a research system for evaluating and improving AI tutors before they are tested with real lear…

KV Cache 让一次推理不用重算历史 token，但多轮对话每次都把 10 万字的 system prompt 重发一遍——前缀缓存回答的是「相同前缀算一遍就够，能不能跨请求共享」。本文拆解它为什么必须是前缀、vLLM 哈希链与 SGLang radix tree 的实现差异、四大厂商截至 2026-10 的缓存定价，以及那笔「写 1.25 倍、读 0.1 倍」的账什么时候是赚的、什么时候反亏 25%。

据报道，OpenAI Safety Systems团队内负责撰写模型system card等公开安全材料的安全透明度负责人David Robinson已离职，离职原因未公开，同时有三名员工因泄密被开除。

Last week, TypeSafe AI released Jev, its first **System One model**. Founder Diogo Almeida previously worked at OpenAI o…

Sep 27, 2026 Researchers from Stanford and Caltech have built HomeBody,** a system that lets a Unitree G1 robot autonomo…

Optimizing deployed AI systems increasingly amounts to editing prompts, skills, harnesses, and code rather than model we…

Tool-calling agents have become central to enterprise AI, yet training and evaluating them at scale remains severely con…

An AI chatbot can give you a bad answer. An AI system controlling an aircraft, vehicle, or robot can create an entirely …

1955 年，乐高推出了 System in Play 玩具平台，要点只有一条：任何一块积木都能和其他积木连接。不同套组的积木由此可以自由组合，拼出设计师从未想到的作品。 也许，这正是「技术平台」的魅力所在—— 只要准备好基础能力、定义连接…

A multi-agent system can reduce latency on complex tasks by executing work concurrently. Several pioneering harness fram…

一份可直接照做的 Jev 实操手册（2026-09-22）：从 state/questions/answers 三件套与 Choice/Score/Noul 三种原语讲起，给出 state 设计、置信度三档路由与阈值标定，Speculative fan-out 与 Composite scoring 两大模式，四个生产级配方（客服分诊 / Agent 工具护栏 / 模型路由 / 上下文压缩），并逐条拆解官方披露的九类失败模式，附成本、限流、版本管理与一页速查表。

Degradations vary widely across images, so a practical restoration system has to handle many degradation types with one …

TypeSafe AI 的 Jev 全面开放，注册即得 5 美元额度（约 1.2 亿输入 Token），输出 Token 永久免费。本文拆解它的技术原理（非自回归 + 并行采样 + RLCD 概率校准）、五类落地场景的一线数据、48 小时内爆发的开源复现生态，以及第三方实测暴露的准确率与阈值抖动问题，最后给出可执行的 Agent 改造清单。

Evaluating an AI system requires disaggregated assessment, as performance varies across domains such as benchmark task t…

System performance, efficient infrastructure scaling and continuous software optimization are key levers that determine …

We present TAPe+ML v3, a compact computer vision system based on TAPe (Theory of Active Perception), a structured repres…

In this work, we present ZGCM-1, a fully open 7B dense foundation model trained from scratch with extreme data, system, …

In hybrid forecasting, a language model is often one of several available signals. A system may already have a market, c…

IT之家 9 月 9 日消息，科技媒体 NotebookCheck 昨日（9 月 8 日）发布博文，报道称奥之心（OM System）新款 PEN 复古无反相机已在美国亚马逊平台开启预售（IT之家发稿前已下架）， 套机售价为 1,299 美…

We present Miles v0.1, a full-stack, production-ready system for frontier post-training. Building upon the clean design …

Recursive self-improvement (RSI) requires a concrete mechanism through which an AI system observes its capabilities and …

Agent harnesses (the system prompt, tool set, execution hooks, and context-management scaffolding around a model) are a …

In systems built on Robot Operating System 2 (ROS 2) and using Data Distribution Service (DDS), a single network-impaire…

We present Srijika, a system for producing installable OpenType fonts for nine Brahmic scripts: Devanagari, Tamil, Benga…

Long-video language models cannot look at every frame: an hour sampled once per second is 3,600 images, and a system kee…

We present a systems-security case study of a two-node split-LLM training system whose privacy evaluation passed while l…

长上下文不是把窗口开大就行：KV Cache 随序列长度线性膨胀，很快压过模型权重成为显存大头。本篇拆解 KV Cache 的来历与账单、多头注意力的三代瘦身方案（MQA/GQA/MLA）、稀疏与线性注意力的路线，以及「上下文长度」宣传数字背后的真实约束。

「这条 SQL 为什么慢」是后端工程师的高频问题，答案往往藏在索引里：B+ 树的结构决定了索引的行为、覆盖索引可以避免回表、最左前缀原则决定了联合索引的可用性。本文讲清 MySQL 索引的原理与优化实践。

「这个应用每月的模型账单会是多少」——回答这个问题需要真正理解 token 计费的口径：输入输出分价、上下文累积、缓存折扣、以及隐藏的成本大头。本篇拆解计费的完整口径、一笔真实对话的成本拆解、以及按效排序的成本优化清单——把「token 焦虑」变成「token 预算学」。

服务器重启后服务要多久恢复？答案藏在启动链路里：固件（UEFI）→ 引导加载（GRUB）→ 内核初始化 → initramfs 挂根 → PID 1（systemd）拉起服务单元。本文逐段拆解每一步做了什么、什么会慢、什么会失败，讲清 initramfs 的角色、systemd 的并行启动模型与依赖图，以及云服务器特有的 virtio/云初始化环节——「重启依赖」从此有全景图。

同一张 GPU，用 naive 推理脚本与用 vLLM，吞吐可以差 10 倍以上——差距的来源是两项工程创新：PagedAttention（显存的分页管理）与连续批处理（请求的动态编排）。本篇拆解这两项技术的机制、vLLM 生态的位置、以及「什么时候该用推理引擎、什么时候杀鸡不必用牛刀」。