Longyue Wang

🌍Our @AI_AlibabaInt has 10 papers accepted by #ICLR2026 and #ICASSP2026!🔥

@grok @erchenlu1 @grok 大学小语种老师，在当下应该何去何从？比如葡萄牙语

被AI/自动化淘汰的职业（如司机、客服、简单编程、数据录入等）人员去向： 好的： - 转行AI相关：提示工程师、数据标注师、AI伦理审核。 - 进入人文领域：教育、护理、创意设计、咨询（人际互动优势）。 - 创业/自由职业：利用经验开网店、技能分享平台。 坏的： - 短期失业，收入断崖式下降，依赖救济。 - 陷入低薪零工经济（如外卖、清洁），不稳定且无保障。 - 适应困难导致心理压力、家庭问题，甚至社会边缘化。 核心：主动学习新技能+政策再培训，能大大提升正面概率。变化是常态，适应者胜。

@alexxubyte @grok 帮我总结

@HuggingModels @grok 帮我看一下技术先进性

Meet LightOnOCR-1B-1025: a game-changing vision-language model that reads documents like a human. It doesn't just extract text, it understands PDFs, tables, and forms with context. This is next-level document AI.

@0x0SojalSec @grok 帮我解释这个工作的核心

RAG AI Agents to process entire codebases and documentation without context limits. It uses a mixture of Experts with Sparse attention to achieve near infinite context in LLMs - github.com/MoonshotAI/MoBA

Haven't gotten around to writing in a bit, here's a short blog on my thoughts since releasing RLMs on the state of AI research. A stronger belief I hold is that future LMs will be scaffolds, and that current LMs are already far more capable than we use them for!

2026 年的套利窗口已经打开。 它还能持续多久？

@AnthropicAI @grok 帮我解释每个行业并实际例子

Software engineering makes up ~50% of agentic tool calls on our API, but we see emerging use in other industries. As the frontier of risk and autonomy expands, post-deployment monitoring becomes essential. We encourage other model developers to extend this research.

New Anthropic research: Measuring AI agent autonomy in practice. We analyzed millions of interactions across Claude Code and our API to understand how much autonomy people grant to agents, where they’re deployed, and what risks they may pose. Read more: anthropic.com/research/measu…

@_vmlops @grok 找个比较正式的中文版看看

Anthropic dropped a 33-page guide on Claude Skills...And this changes how serious teams build AI workflows A Claude Skill is basically a reusable workflow in a folder. One SKILL.md file teaches Claude exactly how you want tasks done consistently every time The real insight isn’t Skills....It’s how to design them properly: • Build micro-skills, not monoliths • Keep instructions short and decisive • Move heavy context into references and assets • Always refine generated Skills manually • Connect Skills to tools via MCP and hooks That’s when AI stops being a chatbot… and starts becoming a system Link - platform.claude.com/docs/en/agents… drive.google.com/file/d/1RR4zKK…

a16z的Bryan Kim刚发了一篇文章，一句话说清楚了AI广告化的必然性： ChatGPT有8亿周活用户，但付费率只有5-10%。 剩下那90%怎么变现？ 答案从互联网诞生第一天就写好了🧵

@benn_huang @grok 分析下这个故事背后真实盈利可能

一周前让 openclaw 去玩 polymarket 然后就忘了，好家伙今天给我发消息已经累计盈利9.8%了…… 我的 skill 是让 openclaw 自己读书总结的，给 AI 的书单：

@FinanceYF5 @Hartdrawss @grok 给我Anthropic指南链接

以上就是全部，原作者 @Hartdrawss 如果您喜欢这个主题： 1.关注我（@FinanceYF5） 2. 点赞+转发下面第一条帖子 x.com/FinanceYF5/sta…

Anthropic 发布了一份 23 页指南，讲解如何用 Claude Code 做几乎所有事情。 以下是完整拆解： 1/ 增长营销（一个人的团队）

黑神话制作人冯骥对于 Seedance 2.0 的评价和判断 到底是做内容的，判断非常准确，这个视频模型对于现有的内容分发、生产、消费体系的影响将会非常大。 我甚至认为字节高层自己都不完全清楚推出这个模型对于他们自己生态的影响。 如果他们确实很清楚的知道，并且依然决定推出的话，我只能说确实字节还是中国最有创新能力和互联网精神的公司。

Camera Language is ready!

@femke_plantinga How about vs. Skills

Confused about MCP and Function Calling in 2026? 9 months ago, we were still debating whether Model Context Protocol (MCP) was just a passing trend or a competitor to Function Calling. Today, we know they are the perfect architectural pair. If you are building agents, here is the updated breakdown: 𝗪𝗵𝗮𝘁 𝗙𝘂𝗻𝗰𝘁𝗶𝗼𝗻 𝗖𝗮𝗹𝗹𝗶𝗻𝗴 𝗱𝗼𝗲𝘀: 1. It helps the LLM decide when an external tool is needed 2. It defines the parameters and schema for a specific task 3. It is usually locked to a single application or specific model provider 𝗪𝗵𝗮𝘁 𝗠𝗖𝗣 𝗱𝗼𝗲𝘀: 1. It standardizes how tools are discovered and served across any system 2. It creates a universal language, so you don't have to rewrite connectors for every new model 3. It enables a plug-and-play architecture where tools are decoupled from the LLM 4. It allows you to build a tool once and have it work everywhere 𝗦𝗼𝗺𝗲 𝘁𝗵𝗼𝘂𝗴𝗵𝘁𝘀: → 𝘞𝘪𝘵𝘩𝘰𝘶𝘵 MCP, you are basically stuck manually wiring every new tool to every new model → 𝘞𝘪𝘵𝘩 MCP, you integrate the server once, and the LLM handles the rest via a standard protocol → You could say that Function calling is the request, where MCP is the interface Standardizing on MCP lets devs stop reinventing hosting patterns and start building more robust agentic workflows. Have you already moved to an MCP-first architecture, or are you planning to do so this year? Lmk in the comments!

@WuMinghao_nlp @Alibaba_Qwen nice!

Personal Update: I'm excited to share that I’ve joined the @Alibaba_Qwen as a Research Scientist! 🎉 I’ll be based in Beijing and continuing my work on multilinguality and large language models. #AI #LLM #Multilinguality #Alibaba #Qwen

Woke up after I used Clawdbot..

Congrats to our team! 🎉5 papers accepted by #ICLR2026 Check out our research lineup below! @iclr_conf