“纯粹结构驱动”推理，是一种不接受过多“先验知识”（即使是概率分布），只接受来自环境的“及时感知”，依靠形式逻辑结构+现实反馈验证，不断地自构命题与验证 这是一种哲学和技术混合的“实验”

the most ambitious and high potential project I have ever done in my life. Will be giving out 5 invite code in this thread

there's a quiet shift happening in how we design software. we're moving from UX to AX (agentic experience). traditional UX is screen-centric. you tap a button, product reacts, job done. every session starts from zero. designers pre-plan every path with hard-coded flows. users

I literally made myself cry when I finished writing this article

Currently in AI coding, people tend to think success = how fast it can build something that looks great, but this is actually not agentic, the agentic success should be = how much autonomy you hand over

试了一下 MyShell 新推出的 ShellAgent 2.0 这个有意思的点是，没有前端页面。 都在一个 Chatbot 里面用通用组件交互，只需要专注 Agent 流程本身就好 用户也不用管展示，只说功能逻辑就行 我用他做了几个好玩的 Agent，👇下面是详细介绍：

We must break free from the assumption that tomorrow’s software is just today’s GUI web app mass-produced by AI; its very form and user experience need a radical re-imagining.

Show Not Tell

🚨BREAKING: this AI agent tells you how to manage your money for a trip better than any travel agent. it booked flights, picked the best cities, built a full itinerary, broke down daily costs, and even found visa tips and ways to save $700+ here's how:

我发现很多人在初学某项知识的时候很容易陷入“定义”/“概念”的分类之争不能自拔，非常教条主义。 比如做ai 产品到现在还在争论agent 和workflow 的区别，对某个产品到底是算agent 还是workflow 争得面红耳赤。一些（自认为）读懂了bitter lesson 的人恨不得把任何workflow

方向明确，任重道远

最近在看sft 和rl 相关内容的时候意识到的： 好的老师做示范（sft），最好的老师只负责出题（rl）

AI素养，或者说AI驾驭能力，很难仅仅通过学习的方式来获得。太需要实践了，而且理论性的东西过时太快。 例如提示词工程，如今已经要让位于上下文工程。但是如果之前缺乏提示词优化的实践经验，理解上下文工程的一些基本概念都会比较困难。

落后的观念才是恐惧的根源。

x.com/i/article/1951…

本次产品宣传看到了一定级别的传播效应，更宝贵的是学习到了很多自媒体们是如何测试各种产品的了。 虽然都说自己是AI博主，有的博主是懂各个产品可能有自己垂类要打的方向，然后在你宣传的方向上重点测试。而大部分则是拿到产品之后一股脑的把测试其他产品的提示词直接灌进去。 一个做chatui app

要开始为了AGI 时代到来之前养成习惯： 能用算力解决的事情尽量用算力，解决不了的再用人力

i feel like we're going to make telegram great again, and i'm serious about this, the internal testing results led to some wild imaginations about its potential

算力是新的杠杆（而且可能是最好的）

x.com/i/article/1954…