from apps to material software used to be something you opened an app was a room with walls: calendar here, notes there, music there, work there. each one had its own logic, buttons, its own little kingdom. the user moved between kingdoms, carrying context in their head but ai starts to break the walls software becomes less like a destination and more like material. something you shape, combine, stretch, ask, remix, and leave behind as traces. a document can become an app. a conversation can become a workflow. a song can become a memory. a task can become an agent. the boundary between using and making gets blurry the old model was: choose the right tool for each task the new model is: express the shape of the thing you want, then refine it with the system you built this changes the role of the interface. ui is no longer only fixed views for fixed functions. it becomes a surface where intent turns into structure. the best interfaces will feel less like menus and more like clay – responsive, persistent, inspectable, and alive apps won’t disappear. rooms are still useful. but the deeper shift is that software stops being a set of sealed containers and becomes a medium people can think through like paper, but executable like language, but spatial like memory, but programmable software stops being something only programmers make it becomes material anyone can shape

Today's Training Data episode takes us BTS on the infrastructure challenges required to do large RL runs at scale, featuring @ellev3n11 (Composer Lead at @cursor_ai) and @dzhulgakov (Co-Founder at @FireworksAI_HQ). The Cursor team trained Composer 2 on Fireworks by starting with a strong base model (Kimi 2.5) and performing large-scale mid-training on code tokens and web data to learn common patterns and libraries, followed by a large-scale Reinforcement Learning run to learn how to navigate the Cursor harness, call tools, and write correct code. Today's episode dives into the systems and infrastructure challenges of making that large RL run happening, and there were many (!!), from numerical mismatch to global distribution to synchronizing rollouts across asynchronous pipelines to keeping track of expert activation across runs and more. Extremely nerdy in-the-weeds challenges that Federico and Dima were delighted to nerd out on together :) Beyond RL infra, we also discussed Online vs Simulated rollouts, self-summarization for long-horizon agents, environment design ("the most powerful RL environment is the product itself"), and other technical nuggets. PS: We filmed this episode before the SpaceX news, while the Cursor team was still compute-constrained. While Cursor now has *all* the flops, the takeaways and hurdles crossed ring true for any serious application-level company that is racing to post-train their own models. I believe that more serious application companies will go the way of Cursor and post-train their own models. 00:00 Introduction 00:53 Why Cursor Trained Composer 2 04:55 Specialization vs Bitter Lesson 06:16 Composer 2 Training Recipe 16:32 Scaling RL Infrastructure Globally 23:32 Floating Point Drift 25:11 MoE Sensitivity Explained 26:25 Router Replay Fix 27:19 Real Time RL Loop 31:49 Long Horizon Agents 34:29 Why RL Everywhere 37:34 LLM as Judge Rewards 39:14 RL in Hard Domains 40:13 Build Your Own Environments 44:34 Closing Thoughts

本周播客：人类和 AI Agent 的最佳配合方式，还没被发明｜对谈 @paperboy_ai y 【视频播客】 youtu.be/Kmpj5KUfIew?si=yWTv… 🚥 本周，十字路口的嘉宾是 Paperboy（paperboy.com）团队。John Yang 21 岁，CEO。Jett Chen 19 岁，CMU 大一在读，同时是 founding engineer。Paperboy 团队 12 人，10 位工程师，融资 470 万美金。 John 认为：人类和 AI Agent 配合工作的最佳方式，很可能还没被发明出来。虽然已经有了 Claude Code、Codex、Manus、OpenClaw，但它们本质上都是 session-based prompt-based。用户需要打开一个窗口，输入 prompt，等完成，关掉。下次从零开始。 Paperboy 正在尝试找到一种更自然、更连续、更可协作的 Agent 界面与记忆结构——Agent 应该通过观察你用电脑来自己学习，用 IM 而不是 session 来组织对话，主动找你，而不是等你 prompt。

Imagine every pixel on your screen, streamed live directly from a model. No HTML, no layout engine, no code. Just exactly what you want to see. @eddiejiao_obj, @drewocarr and I built a prototype to see how this could actually work, and set out to make it real. We're calling it Flipbook. (1/5)