Latest

Every published item, in order · 55859 items
Nov 12
How-toBuildersOfficial12:11
Chinese content一篇Claude Code实操笔记:用skill系统从vibe coding走向正规工程

写代码的人看这个:作者把Claude Code的skill系统玩明白了,连安装、创建自定义skill的命令都给出来了,还顺手写了测试、部署相关的模板。附赠的省钱配置也挺实在,每天3美元跑agent,比很多人烧几十美元省多了。

Primary source歸藏(guizang.ai)@op7418Source
IndustryBuildersOfficial12:11
Chinese contentGoogle Research 与临床医生合作,让医疗模型学会向患者提问

Google Research 分享了一个医疗 AI 的思路:模型不能只答对,还得像医生一样会追问患者情况。他们和临床医生一起训练模型理解患者上下文。做医疗 AI 或对医患对话建模感兴趣的人可以看看原文链接。

Primary sourceGoogle Research@GoogleResearchSource
ModelsBuildersSelected12:09
Ant LingBo opens-source LingBot-Touch tactile foundation model

A unified tactile representation that transfers across sensor types and hand hardware removes one of the biggest integration costs in robot manipulation research. Teams building dexterous manipulation with VLA or diffusion policies can plug it in and benchmark gains on their own stacks instead of collecting tactile data from scratch. As an open-source Chinese entrant, it gives developers in Southeast Asia and beyond a concrete alternative to proprietary embodied-AI stacks when choosing perception components.

IT之家Source
ProductsBuildersOfficial12:09
Chinese contentAmazon Bedrock AgentCore 推出 Web Search API,一次签名调用即可联网搜索

AWS 给 Bedrock AgentCore 加了个 Web Search API,一行签名调用就能让程序或智能体搜网页,不用自己搭网关。做 Agent 或数据管道的开发者可以直接在 SDK 和 CLI 里接上用,省掉不少基建活。

Primary sourceAWS Machine Learning Blog@Shubham DivekarSource
PapersBuildersMulti-sourceSelected12:07
NVIDIA Déjà View turns multi-angle photos into 3D geometry

Weight-tied looping shows parameter efficiency, not scale, can dominate in 3D vision tasks, which shifts how small labs should think about model budgets. Teams doing photogrammetry, robotics perception or 3D content generation can deploy it on modest hardware and tune loop count to their compute ceiling. Code is available via NVIDIA's research page, making it directly testable against larger pose-estimation baselines.

Event
Official accountNVIDIA AI@NVIDIAAISource
How-toBuildersSelected12:07
Together AI: Match Inference Bottlenecks to Your Traffic Shape

The practical shift here is treating provider benchmarks as workload-specific rather than universal — a provider that wins on decode speed can lose badly on prefill-heavy or agent workloads. Engineering teams evaluating inference vendors or deciding between self-hosting and APIs should benchmark with traces from their own traffic before committing. Concrete comparison: run the same prompt set against at least two providers and measure the metric matching your traffic shape, not aggregate throughput.

Official accountTogether AI@togethercomputeSource
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
ModelsUS vs ChinaMulti-source12:06
Chinese contentMeta 发布 Muse Ember 小模型,专攻智能体重复性任务

Meta 新出的 Muse Ember 走小而专的路线,专吃工具调用和路由这类高频低难度活,单位成本上压过 Claude Haiku 5.5 和 GPT-6 Luna。在跑大规模智能体流水线的团队可以算算账,但写代码还是得换别的模型。

Event
Official accountDecoder@Matthias BastianSource
Chinese contentMatt Shumer 征询是否公开内测其智能体游戏

Matt Shumer 想放出一个还在开发中的游戏 build,让玩家试玩并给他做的智能体提反馈。如果一直在关注他的智能体产品,这是一个直接参与早期测试的机会。不过帖子只有一句话,没有产品截图或功能介绍,信息量很少。

Event
Matt Shumer@mattshumer_Source
Chinese content苹果将在土耳其设7处培训设施,含美国以外首个先进制造中心

苹果在土耳其的布局不只是开几所学院,而是把美国本土的先进制造培训模式首次复制到海外,中小企业能免费学到预测性维护和机器人技术。开发者还能通过 9 个月免费课程学到 AI 相关技能,东南亚和中东地区的开发者生态值得关注这类似模式是否会扩展。

IT之家Source
ModelsUS vs ChinaSelected12:03
Baidu ERNIE 5.5 Flash Preview Enters LMArena Battle Mode

The sparse 14B-active design plus 200+ tokens/sec decoding makes this one of the more inference-efficient large open-weight models to watch, and independent Arena battles will give a less curated signal than vendor benchmarks. Developers evaluating coding agents or self-hosting options should compare it against DeepSeek and Qwen once weights drop by 25 November. Baidu's continued open-weight push also matters for teams in China and Southeast Asia seeking alternatives to US-hosted APIs.

lmarena.ai@lmarena_aiSource
How-toBuildersMulti-source71°12:03
Chinese content斯坦福新开 CS329Z 智能体工程课程,第一讲公开:智能体系统详解

斯坦福这学期新开的智能体课免费放出来了,第一讲是 Diyi Yang 主讲,配中英字幕。她用 ReAct 论文、OpenClaw 删邮件这些例子把智能体的循环结构和记忆设计讲得很透,想做智能体的开发者可以直接当入门课看。

Event
宝玉@doteySource
ProductsBuildersOfficial12:03
Chinese contentOpenAI Codex 现已支持 Amazon Bedrock 作为模型后端

企业开发团队想用 OpenAI Codex 又怕代码数据出不了 AWS 的,这篇官方指南把访问、审计和计费怎么配都讲清楚了。已经在用 Bedrock 管 Claude 的团队照着配就行,不用给 OpenAI 另开一条通道。

Event
Primary sourceAWS Machine Learning Blog@Shivi BhatiaSource
IndustryBuildersMulti-source12:03
Chinese contentNvidia 与自动驾驶公司 Autobrains 扩大合作,采用 Alpamayo 2 Super 训练驾驶智能体

以色列自动驾驶公司 Autobrains 拿 Nvidia 的 Alpamayo 2 Super 来训练针对具体路况的驾驶智能体,双方是老搭档,这次算是加深绑定。做自动驾驶或具身智能的团队可以看看这种大模型场景化训练的路子怎么落地。

Event
Tech in Asia@Diya LalSource
ModelsBuildersSelected12:03
SuperMimic trains Unitree G1 whole-body motion with a differentiable world model

Model-based training could cut the compute and sim-hours that RL pipelines demand, but the sim-to-real gap on contact-rich edge cases becomes the new bottleneck. Teams building motion-tracking skills for humanoids can test SuperMimic against their existing DreamerV3-style or PPO pipelines using the open GitHub code. Anyone deploying G1-class robots in unstructured environments should weigh the noted failure modes on collisions and slippage before relying on it.

Gorden Sun@Gorden_SunSource
Chinese contentNous Research 的 Hermes Agent Plugin 黑客松收到近 1000 份参赛作品

Nous Research 的 Hermes 插件黑客松收了近 1000 份作品,评审正在选出 top 10,获奖者一周内公布。想做智能体插件开发的人可以关注公布结果,看看社区里哪些插件思路被选中,比看官方文档更直观。

Official accountNous Research@NousResearchSource
Chinese contentFrançois Chollet 谈教育 AI:不该替学生拿答案,Gakku 获融资

Chollet 本人都出来站台了:他批评现在教育 AI 一味给答案,夸 Gakku 是教学生自己思考的那类产品,还刚拿了融资。家里有娃在用 AI 写作业的,或者做教育产品的开发者,都该看看这个思路差在哪。

Official accountFrançois Chollet@fcholletSource
ModelsUS vs ChinaSelected12:01
Claude Fable 5.2 fails without tools on VIALS-AA

Model rankings can flip entirely depending on whether the evaluation harness includes tools, so leaderboard positions measured one-shot may mislead procurement decisions. Teams building scientific or image-heavy agent pipelines should benchmark cheap models like Claude Haiku 5.5 inside their own sandbox before paying flagship rates. The cost gap, US$0.07 versus US$4.20 per task, makes the harness-versus-model trade-off a concrete budgeting question rather than an academic one.

Event
Official accountArtificial Analysis@ArtificialAnlysSource
Chinese contentGPT Image 2.5 Sunburst 对比 Seedream 5.5:连续 15 轮编辑实测

有人真拿两个文生图模型连改 15 轮,每轮基于上张输出,结果 Seedream 5.5 到第 10 轮画面还没崩,GPT 版已经橙化锐化过头。经常改稿改到第 N 版的设计师和运营可以看看,毕竟每张才 0.02 美元。

kimmonismus@kimmonismusSource
Chinese content一条推文引发讨论:AI Native 产品与 Super Agent 的终极形态是什么

这位作者抛出一个产品形态判断题:AI Native 产品和 Super Agent 长什么样才算到底。做智能体或 AI 应用的开发者可以对照一下自己的产品路线,看看是在旧壳里塞模型能力,还是真正按模型能力重构产品。

berryxia@berryxiaSource
ModelsUS vs ChinaSelected12:00
GLM-5.3 runs on SambaCloud at 200+ tokens per second

Token efficiency per task is the metric here — fewer output tokens means lower API bills and faster agent loops, which matters more than raw benchmark scores for production coding agents. Teams running Claude Code or Codex CLI can test the setup by swapping a single model parameter. Compared with closed frontier options, GLM-5.3 offers an open-weight alternative served on dedicated SambaNova hardware rather than shared GPU clusters.

Event
SambaNova@SambaNovaAISource
Chinese contentHelsing 发布 Clarity:用 AI 智能体为武器系统生成目标建议

这家德国防务公司把传感器数据交给智能体自动整理成打击目标清单,但坚持每个决定写上真人名字。对比美军同类系统更强调欧洲自主可控。做安全、军事 AI 或关心自动化杀伤链的人值得细看里面的监督机制。

Official accountDecoder@Maximilian SchreinerSource

Only the latest 2000 items are listed; for older ones see the daily archive