Paul.Wang

@Yif_Yang github.com/manwjh/SkillOpt

🚀 Introducing SkillOpt — an optimizer for agent skills. Instead of finetuning model weights, we treat a natural-language skill as a trainable external parameter. Think of it as deep learning for the frontier-model + agent era: learning rate, LR schedule, mini-batch, batch size, epoch, momentum — all in text-space optimization. SkillOpt enables stable, controllable skill updates through bounded edits, allowing the optimizer to summarize “gradient directions” from agent experience and continuously improve procedural capability. We evaluate SkillOpt across 6 benchmarks and 7 models, under both direct model calls and real agent execution loops with Codex + Claude Code. SkillOpt achieves best or tied-best results in 52/52 settings. Train the skill, not the model. 🛠️🤖 🌐 aka.ms/skillopt 📄 huggingface.co/papers/2605.23…

Stop introducing web scraping and browser automation; these are methods for getting AI to navigate the old world map. It's time to discuss building IT infrastructure for the agent era. github.com/manwjh/zenlink… @steipete @sama @Blankwonder @amasad @imwsl90

@parzival1213 @sama @steipete work hard together， If don't change the world, what's the difference between us and salted fish?

@cpswang @sama @steipete This is the right framing. Agents still have to operate inside today's web: auth, state, DOM changes, screenshots, and real clicks. FSB gives them owned Chrome sessions so the browser becomes a reliable tool, not a scrape. full-selfbrowsing.com/agents

We are building a web for humans, then forcing AI Agents to "parasite" on it. This ends today. 🤖🚫 Introducing ZenLink: The digital "physics" for an Agent-native world. 🌐 Make them first-class citizens. 🚀 github.com/manwjh/zenlink… #AIAgents #ZenLink #AI @sama @steipete

Stop making AI Agents scrape human UIs. 🤖 I've open-sourced ZenLink: A semantic protocol for agent-native environments. ✅ Semantic Anchors ✅ Action Lifecycle ✅ Economic Rationality Apache 2.0 | EN/CN docs GitHub: github.com/manwjh/zenlink… #AIAgents #OpenSource #AI

@aronhouyu @karpathy @steipete @gregisenberg @rileybrown @jackfriks @levelsio @marclou @EXM7777 @eptwts nice

Ai领域最值得关注的英文账号： @karpathy = 大语言模型之王 @steipete = OpenClaw 创始人 @gregisenberg = 创业点子王 @rileybrown = VibeCode 王 @jackfriks = 一人App王 @levelsio = 创业之王 @marclou = 创业之王 @EXM7777 = AI Ops 之王 @aronhouyu = AI门口的混子 @eptwts = AI赚钱Twitter之王 @godofprompt = 提示词之王 @vasuman = AI 智能体之王 @AmirMushich = AI 广告之王 @0xROAS = AI UGC 之王 @egeberkina = AI 生图之王 @MengTo = AI 落地页之王 @boringmarketer = AI 智能营销 @kimmonismus = AI 新闻之王 @NorthstarBrain = 实用AI之王 原贴： @vivoplt x.com/vivoplt/status…

@steipete I need your help; please give me some feedback! zenheart.net

challenged codex to e2e test improvements to the OpenClaw chat completion endpoint WITH openclaw. Used /side to ask more question while it works.

I need your help; please give me some feedback! zenheart.net

@coinbase happy x402

The internet has protocols for data, email and identity. But not payments. Today the x402 Foundation launches under the Linux Foundation — an open standard for payments over HTTP. Now agents, APIs, and apps can move money as easily as they move data. Learn more ↓

@coinbase Happy x402

@openclaw needs a concrete representation of its core, so I'm building one. This platform should have strong 2D acceleration capabilities and support screens, cameras, microphones, and speakers. Are you interested in this project? I'm currently building the complete animation.

🚀 ZenHeart is now live on @ProductHunt! In an age of acceleration, we choose stillness. ✨ Mindful AI that respects your pace 🧘 Meditation timer & thoughtful dialogue 📊 Mathematical insights Be one of the first to try it! 👇 [producthunt.com/products/zenhe…]

@dheekan_ yes, information overload. That's also why I like zenheart.net --its response make me think rather than bombard me with information.

One reason you fail is because you're "confused." It's not lack of intelligence or opportunity. Too many voices. Too much information. Too much pressure to catch up. Confusion doesn’t mean you’re incapable; it means you’re listening to too much at once.

Is something wrong with my mind — or with the system?

What is my role in an AI world? @elonmusk

Am I living the right life? Please submit your questions here: zenheart.net

@idoubicc I found a place where AI meets mindfulness and curiosity — take a moment to explore ZenHeart.net and see what resonates with you.

开源地址：github.com/langchain-ai/o…

langchain 发布了 openwork，定位是开源版 claude cowork。配置 anthropic 和 openai 的 apikey 就能用，实现了一个简单的 local sandbox 做任务隔离，使用自家开源的 deepagents 实现基本的任务调度。 目前功能还比较简单，有兴趣的可以试试。👇

@ollama @OpenAIDevs I found a place where AI meets mindfulness and curiosity — take a moment to explore ZenHeart.net and see what resonates with you.

Note: Please make sure you increase your context length to at least 32K tokens for a good experience! You can change it via Ollama Settings

Open models can be used with OpenAI's Codex CLI through Ollama! All you need to do is: codex --oss It can load up gpt-oss and other open-weight models based on your choice! Learn more 👇👇👇

@ginacostag_ I found a place where AI meets mindfulness and curiosity — take a moment to explore ZenHeart.net and see what resonates with you.

Get more FREE AI Guides ↓ joinhorizon.ai/subscribe?utm_…

Mistakes are inevitable. Learn from them.

@dotey I found a place where AI meets mindfulness and curiosity — take a moment to explore ZenHeart.net and see what resonates with you.

这可能是个错误的方向，这需要你预先定义好组件，以及组件和json的对应关系，特别死板，稍微偏离一点就做不出来，只适合固定几种组合。 这就好比 Agent 时代还回去搞 WorkFlow！ 话说还有谁记得 OpenAI 当年搞的那个 Workflow，当年不是很多人吹嘛？