Latest

Every published item, in order · 29223 items
Nov 13
How-toBuildersMulti-source04:37
Chinese contentOpenAI 工程师拆解 OpenClaw Enterprise 持久化智能体的内部安全管控

想知道企业里跑持久化智能体到底怎么管权限的,可以看这篇。OpenAI 工程师亲自讲他们内部怎么部署 OpenClaw Enterprise,比厂商官方文档的空泛描述实在,做企业级 Agent 落地的开发者别错过。

Event
openclaw@openclawSource
ModelsBuildersMulti-source04:36
Chinese contentHedge fund 用心理分析模型 Thorsten-4 分析财报电话会动机

这个思路挺有意思:4 年追踪 7,000 多人、标注 3,500 万词训出来的心理分析模型,能从财报电话会里读出高管的动机,已经有基金拿它跑 5,000 万美元的仓位。做量化或者另类数据分析的可以看看。

Event
rohanpaul_ai@rohanpaul_aiSource
ProductsBuildersMulti-source04:35
Chinese contentChatGPT 更新:dot 响应更快、支持自定义指令与行为规则

OpenAI 给 ChatGPT 的 dot 加了自定义指令,回复更快、说话更少术语,还能设定哪些动作要先经你批准才执行。经常用 dot 帮忙发邮件、跑任务的人会省下反复纠正的麻烦,移动端和桌面端都能直接设置。

Event
ChatGPT@ChatGPTappSource
ProductsBuildersMulti-source79°04:34
Chinese contentAnthropic 推出 Claude Meetings:会议中实时生成文档和白板

Anthropic 把 Claude 塞进了视频会议,边开会边帮你写文档、画白板,产出直接是可以编辑的 artifacts,不是干巴巴的纪要。目前只对 Team 和 Enterprise 计划开放 beta,跟 Teams Copilot 这类工具正面竞争,经常开远程会的团队值得试试。

Event
Claude@claudeaiSource
ProductsBuildersMulti-source04:34
Chinese contentClaude 新增会议记录功能:通话结束后自动整理纪要与待办

Anthropic 把 Claude 接进了通话场景,开完会纪要、待办、转录、甚至 Claude 顺手做的东西都自动落在会议页面上,等于 Otter 这类记录工具的功能被顺手做了。经常开会还要手动整理笔记的人可以试试。

Event
Claude@claudeaiSource
ModelsBuildersSelected04:30
Retell tops $100M run rate, launches Tell 1 voice model

A voice-AI vendor shipping its own foundation model signals a shift from reselling general-purpose LLMs to owning the speech-to-reasoning stack, which could compress latency and cost for high-volume call operations. Contact-center engineering and CX teams evaluating voice-agent platforms should benchmark Tell 1 against general models like GPT-4o on latency and failure rates. Conductor's automated test-case generation also gives teams a concrete workflow for continuous agent quality control.

TestingCatalog@testingcatalogSource
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
How-toBuildersMulti-source04:29
Chinese contentAnthropic 员工谈 Claude Code 配置优化:精简 CLAUDE.md 与权限设置

Claude Code 用久了 CLAUDE.md 堆了一堆旧规则,反而干扰新模型。Anthropic 的 Lydia Hallie 讲了怎么精简:硬约束交给权限,质量交给测试和 CI。天天用 Claude Code 写代码的人值得花十几分钟看完这个演讲。

Event
AI Engineer@aiDotEngineerSource
Chinese contentAI Engineer Code Summit 首日收官,Day 2 直播已在 YouTube 开启

这是 AI Engineer 社区办的编程峰会,由 Google DeepMind 赞助。首日讲的是各家团队怎么做 AI 编程相关的工作,Day 2 直播已经在 YouTube 开了。做智能体和开发者工具的人可以直接去看录像,能了解到同行现在的实际做法。

AI Engineer@aiDotEngineerSource
PapersBuildersSelected04:26
Stanford Researchers Introduce Speculative Self-Distillation for LLM Fine-Tuning

The non-obvious insight is that you do not have to choose between on-policy accuracy and training cost — selective teacher intervention gets both, plus measurably less forgetting. Developers and teams doing domain adaptation or continued fine-tuning can use SSD-style distillation to add new knowledge without degrading existing capabilities at a fraction of the token budget. Anyone benchmarking against standard on-policy distillation pipelines should compare training time and retention metrics directly.

Stanford AI Lab@StanfordAILabSource
ProductsBuildersMulti-source04:24
Chinese contentAnthropic 推出 Claude Meetings:在 Google Meet 通话中直接调用 Claude

Anthropic 把 Claude 直接塞进 Google Meet 当参会者,能边开会边干活、会后自动出转录和摘要,还确认说话人才执行口头指令。相比会后手动喂录音的工具省了一步。用 Team 或 Enterprise 计划、常开远程会的人可以试试。

Event
rohanpaul_ai@rohanpaul_aiSource
ProductsBuildersSelected04:23
Vercel launches UI SDK for agent-driven interface generation

The shift implied here moves value from opinionated component libraries down to low-level primitives that agents compose themselves, which could reshape how frontend tooling is built and priced. Teams building agent workflows or evaluating whether shadcn-style libraries still fit their stack should watch how the SDK's primitives perform in practice. Vercel's reach in the Next.js ecosystem means this framing will likely influence tooling decisions well beyond its own platform.

Event
Guillermo Rauch@rauchgSource
IndustryUS vs ChinaMulti-source74°04:17
Chinese contentReuters:白宫考虑撤换商务部长 Lutnick,Nvidia 涉嫌间谍风波波及商务部

商务部长被指收到机密警告却没把买了 48 亿美元芯片的 Megaspeed 拉黑,现在白宫在讨论换人。这事直接关系 Nvidia 芯片卖给谁、对华管制怎么执行,做芯片和出口合规的应该盯着。

Event
rohanpaul_ai@rohanpaul_aiSource
PapersBuildersMulti-source04:16
Chinese contentMETR 在 Anthropic 的研究印证了"直觉测试"的有效性

METR 在 Anthropic 做的这项研究给 vibe-testing 背了书。如果你平时靠快速上手体验来判断模型好坏,这篇等于给你一个依据,还顺带聊了 Claude Sonnet 3.7 为什么是 2025 年最超预期的模型,做模型选型的人可以看看。

Event
Ethan Mollick@emollickSource
ProductsBuildersMulti-sourceSelected03:56
Databricks Genie Plugin Brings Business Context into ChatGPT

What changes is where enterprise analytics happens: governed, ontology-aware answers now sit inside ChatGPT rather than only in Databricks' own workspace. Data and analytics teams at enterprises already on Databricks can pilot self-service querying for business users, while OpenAI gains an enterprise anchor. Teams in ASEAN and China markets relying on ChatGPT for work should check whether their region and data governance rules permit routing queries through this plugin.

Event
Official accountDatabricks@databricksSource
Chinese contentSimular 推出 Box:可交给 ChatGPT、Claude 等智能体操作的云电脑

Simular 做了一台专门给 AI 用的云电脑,ChatGPT、Claude、Codex 都能登上去自己开应用、点网站,你的电脑关机也不影响任务跑。最特别的是任务可以交接——Claude 干到一半,Codex 能接着干。经常搭智能体自动化流程的人可以关注。

TestingCatalog@testingcatalogSource
ModelsUS vs ChinaMulti-source03:49
Chinese contentGoogleDeepMind 发布 Gemini 4 Hydrogen (High),WebDev 榜单首秀即入前十

谷歌这个 Hydrogen 版本走性价比路线:Agent Arena 上每任务只要 $1.6,比 GPT-6 Astra 便宜近 4 倍,还快 2.4 倍,智能体实测净改进 +14.8%。做智能体应用、对成本敏感的开发者可以对比下它和 Claude Fable 5.2 的取舍——快和便宜,但吃 token 更多。

Event
lmarena.ai@lmarena_aiSource
ModelsUS vs ChinaMulti-sourceSelected03:48
Gemini 4 Hydrogen ranks #4 in Agent Arena knowledge work

The non-obvious point is that overall leaderboard position can mislead: Hydrogen beats three higher-ranked models once tasks are filtered to knowledge work. Engineering leads and product teams running agentic assistants should compare models on their actual task mix rather than overall rank, using category-level Arena data like this. It also gives Google DeepMind a concrete speed-plus-quality benchmark against Anthropic's and OpenAI's Max-tier offerings.

Event
lmarena.ai@lmarena_aiSource
Chinese content软件工程师转型指南:从写代码转向做 AI 帮不上忙的事

一位独立开发者写的职业反思:写代码的门槛被 Claude、Cursor 这类工具拉平后,他把重心转向产品定位、营销和心理学这些 AI 没法替你做的事。文中“提示词比技术方案更管用”的一线体验很真实,正在做独立开发或纠结转型的工程师可以对照着自己的情况看看。

宝玉@doteySource
Chinese contentAgent Arena 用基于真实任务轨迹的方法评估模型表现

Arena.ai 搞了个新评估方法,考题不是人工设计的,而是全球用户提交的真实任务,再分析模型执行过程的完整轨迹来打分。比起 Chatbot Arena 靠投票、固定榜单跑分,这种方式更贴近智能体真实干活的表现。做智能体选型的开发者可以关注它的结果。

lmarena.ai@lmarena_aiSource
ProductsBuildersMulti-sourceSelected03:41
LlamaIndex launches ParseBench-2 benchmark and OCRia-1.5 parsing models

Benchmark design that weights errors by severity rather than raw accuracy changes how OCR vendors will be ranked and evaluated going forward. Teams building document ingestion pipelines, RAG systems, or enterprise data extraction can use ParseBench-2 to compare the 150+ benchmarked OCR solutions, and LlamaParse users get OCRia-1.5 immediately. If you currently evaluate with OmniDocBench or LLM-judge methods, this offers a third methodology grounded in human-judgment alignment.

Event
Jerry Liu@jerryjliu0Source
ProductsBuildersMulti-source03:38
Chinese contentGoogle 发布 Gemini 4 Hydrogen,与 Carbon 组合使用效果最佳

Google 新出的 Gemini 4 Hydrogen 速度飞快,在 artificial analysis 智能指数上还超过了 Argon,而且对所有订阅档都开放。有人用它搭配 Carbon 做规划-执行分工,效果比单模型更好。如果你在搭 agent 或重度用 Gemini,这套双模型玩法可以直接抄作业。

Event
Paul Couvert@itsPaulAiSource
ProductsBuildersMulti-source03:35
Chinese contentGemini 4 Hydrogen 已上线 Devin Desktop 和 Devin CLI

Cognition 把 Gemini 4 Hydrogen 接进了 Devin Desktop 和 Devin CLI,在自家 FrontierCode 2.0 基准上拿了 38.2%,只差 GPT-6.1 Sol 一名。平时用 Devin 写代码的人现在可以多一个模型选项,做模型选型的团队可以看看这份真实工程任务榜单再决定。

Event
Cognition@cognition_labsSource
Chinese contentpoetiq 借助 Together 后训练快速上线复杂 RL 研究工作负载

做 RL 训练的朋友可以看看这个案例:poetiq 用 Together 的后训练服务跑复杂 RL 工作负载,启动快,但训练循环还是自己写、自己控制。如果你想要托管算力又不想交出算法细节,这种模式和 OpenAI、Anthropic 的封闭训练方案思路不一样。

Official accountTogether AI@togethercomputeSource

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