Latest

Every published item, in order · 29223 items
Nov 13
ModelsBuildersMulti-source00:33
Chinese contentmillionways 发布 Thorsten-4,基于 7000 人四年纵向访谈数据训练

7000 人四年访谈数据训练出来的对话理解模型,验证方式是拿预测对照真实行为,不是刷基准分。做客服机器人、心理支持类智能体的团队可以直接看 demo,思路和主流爬语料训模型完全不同。

Event
kimmonismus@kimmonismusSource
IndustryPolicyMulti-source00:32
Chinese contentAnthropic 在 IPO 招股书中称愿为安全放慢发展速度

这段话出现在上市文件里挺少见——公司直接承认会为了安全接受落后。对比 OpenAI 一路高歌猛进的节奏,Anthropic 把安全写进招股书更像是在给监管和投资者打预防针。关注模型发布节奏和安全策略的人都该看看原文。

Event
kimmonismus@kimmonismusSource
IndustryUS vs ChinaMulti-source81°00:32
Chinese contentAnthropic IPO 路演披露:Claude 三天完成 AMD 芯片适配

Anthropic 上市路演里信息量不小,最硬核的一条是 Claude 三天搞定模型在 AMD 芯片上的适配,以前要几十个工程师干几个月。对比 OpenAI 还没上市的节奏,这份披露很值得细看,想了解 Claude 实际能力的开发者别错过。

Event
kimmonismus@kimmonismusSource
ModelsBuildersSelected00:31
SOV-Transcribe-1 Tops Open Weights Speech-to-Text Rankings

A 4.6B-parameter open model beating proprietary transcription APIs at 2.3% WER shifts the economics of self-hosted speech workloads, since teams no longer pay a premium for accuracy. Developers building voice agents or call analytics should benchmark it against Whisper-based stacks and hosted APIs on their own audio domains. The small active parameter count also makes GPU deployment affordable for smaller teams and regional SaaS builders.

Official accountArtificial Analysis@ArtificialAnlysSource
Chinese contentGoogle 今日再次发布新品

Google 今天又发新东西了,博主 kimmonismus 在 X 上发了条提醒。帖子本身没给细节,但如果你在追 Google 这波密集更新,这条可以当个线索,去官方渠道看看具体发了啥。

kimmonismus@kimmonismusSource
PapersBuildersSelected00:31
Recursive Self-Improvement Applied to Frontier Pre-Training Run

The non-obvious takeaway is that RSI may pay off more in pre-training headroom than in post-training tricks, since a barely fine-tuned base beat fully post-trained rivals. Researchers running large-scale training pipelines and labs weighing agent-assisted data curation should study the loop decomposition and shared context store design. The model being open-weight and sized at 30B MoE makes it directly comparable for local deployment against Qwen3-30B-A3B and Nemotron-3.5-Lightning-30B-A3B.

DAIR.AI@dair_aiSource
ModelsBuildersMulti-sourceSelected00:29
MiniMax H3-Image Tops Artificial Analysis Open-Weight Image Editing Rankings

The notable detail is the Vera Rubin support landing in vLLM-Omni ahead of the hardware's broad rollout, so early adopters can benchmark image-editing serving costs on next-gen GPUs from day one. Teams comparing open-weight image editing models should add H3-Image to their shortlist against FLUX-family alternatives. The 0.87-second 4K generation figure is a useful baseline for latency-sensitive production deployments.

Event
Official accountvLLM@vllm_projectSource
How-toBuildersMulti-source00:28
Chinese content按任务分配模型:Agent 跑任务别全程用最贵的模型

跑 Agent 最费钱的就是全程顶配模型。这篇讲了按任务分流的实操:规划用 Kimi K3.1 Pro 这类强模型,读文件改格式丢给 0.01x 倍率的豆包 2.1 Lite,同样额度能多干好几倍活。经常跑自动化任务的人照着配一遍就能省不少。

Event
Lxfater@lxfaterSource
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
Chinese contentAlibaba Cloud Summit 香港 2026 将设 AI 电影创作者圆桌论坛

11 月 17 日在香港的这场圆桌请了三位真在做 AI 视频的创作者,聊从创意到成片的完整流程。不是厂商讲 PPT,而是讲 agentic 系统怎么进实际制作管线。做短视频和广告内容的团队可以关注。

Official account阿里云 Alibaba Cloud@alibaba_cloudSource
ProductsBuildersMulti-source00:26
Chinese contentQwen4-Max-Preview 上线 Alibaba Cloud AI Token Plan

阿里的新旗舰 Qwen4-Max-Preview 直接塞进了 3 美元首月的订阅档,想提前试旗舰能力的话成本很低。比起按量付费的 OpenAI 和 Anthropic API,这种包月模式对小额用量开发者更友好,适合想测 Qwen 系列的人上手。

Event
Official account阿里云 Alibaba Cloud@alibaba_cloudSource
Chinese contentAlibaba Cloud 参展 HKICC 2026,现场展示 Qwenwork 一体化 AI 生产力平台

Alibaba Cloud 把 Qwenwork 这款智能体生产力平台直接带到 HKICC 2026 现场给香港企业看,说明它把香港当成 agent 工具的重点市场。做知识工作自动化、选型 agent-first 工具的团队可以关注它和 Notion AI、钉钉 AI 这类产品的差异。

Official account阿里云 Alibaba Cloud@alibaba_cloudSource
Chinese content百度称自动驾驶是AI竞争关键,看好Apollo Go前景

百度在自家Baidu World 2026大会上把Apollo Go摆到了AI竞争的核心位置,Robin Li还拿具身智能类比2016年的自动驾驶。如果你关注无人出行和L4路线,这篇能看清百度的四层技术栈打法,和Tesla、Waymo的路数对照着看更有意思。

Official account百度 AI Baidu@Baidu_IncSource
ModelsBuildersMulti-source00:25
Chinese content百度世界 2026:Robin 李彦宏阐述模型与智能体并行演进路线

百度这次把深度思考和微表情识别两款 ERNIE 模型直接开放到千帆平台 API,企业接上就能用。李彦宏讲的思路很有意思:智能体的应用反馈反过来指导模型优化。做企业级应用开发的可以看看千帆上的调用方式。

Event
Official account百度 AI Baidu@Baidu_IncSource
ProductsBuildersSelected00:24
Baidu launches Token Exchange marketplace for AI models

A marketplace layer with fingerprint checks and public credit records could pressure model providers toward verifiable pricing and specs, shifting trust from vendor claims to platform-verified data. Enterprise buyers comparing models across providers get a single venue to evaluate options, useful for teams in ASEAN and China sourcing models without direct vendor relationships. Compare with direct API portals from OpenAI or Alibaba Cloud, which lack third-party verification layers.

Official account百度 AI Baidu@Baidu_IncSource
IndustryUS vs ChinaMulti-source76°00:19
Chinese contentNvidia CFO Colette Kress 曾反对向 OpenAI 提供 2500 亿美元信贷担保

Nvidia 内部原来也不是铁板一块:CFO Kress 拦下了 Jensen 主推的给 OpenAI 的 2500 亿信贷担保,现在她还当上了临时 CEO。想知道 Nvidia 那些投资换订单的生意到底有多激进、谁在踩刹车,这篇值得花十分钟看看。

Event
The Information@theinformationSource
Chinese contentLangChain 发起 State Of Agent Engineering 年度调研,约 5 分钟即可参与

LangChain 在收集今年智能体工程调研的数据,5 分钟就能填完,结果会公开发布。做智能体开发的可以顺手填一份,顺便看看大家现在都在用什么框架和评估方式,等报告出来对照自己的技术栈。填了还能抽周边。

Official accountLangChain@LangChainAISource
IndustryUS vs ChinaMulti-source00:11
Chinese contentAravind Srinivas 发文致敬 Jensen Huang 33 年 Nvidia 生涯

Perplexity 创始人 Srinivas 发长文感谢 Jensen Huang,说他把 Nvidia 从小公司带到美国市值第一,33 年了。如果你关注 GPU 供应链和后续交棒,这条值得收藏,评论区还能看到各家创始人跟进表态。

Event
Aravind Srinivas@AravSrinivasSource
ModelsBuildersMulti-source00:04
Chinese contentmillionwaysai 发布 Thorsten-4 模型,可对对话打 50 项行为心理评分

millionwaysai 做了个叫 Thorsten-4 的模型,读完对话能给 50 项行为心理特质打分,直接接进客服或销售工具判断用户意图和反应。做用户洞察、CRM、销售工具的开发者可以看看,比问卷画像省事不少。

Event
Ate-a-Pi@svpinoSource

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