灵巧手的数据格式： task_info/task_327.json： { "episode_id": 649078, "task_id": 327, "task_name": "Picking items in Supermarket", "init_scene_text": "The robot is in front of the fruit shelf...", "lable_info": { "action_config": [ {"start_frame": 0, "end_frame": 435, "action_text": "Pick up onion from the shelf.", "skill": "Pick"}, {"start_frame": 435, "end_frame": 619, "action_text": "Place onion into the plastic bag.", "skill": "Place"} ] } } 亮点：指令不是整段一个，而是按帧区间切片 + 原子技能(skill)标签分层标注。 这比我们 language_instruction.json 只标一个 segment 更细——它把"拿起→放下"拆成两个子动作各配指令。 这正是高质量示教数据的做法。

1,000,000 rigid bodies. All simulated on the GPU. No GameObject per body. No CPU readback — the CPU never sees a single pose, or even how many got drawn. Bodies, broadphase, solver, culling and the instance stream all live on device. Then I dropped four tornadoes on it.

- Claude = coding. ($20/mo) - Supabase = backend. (Free) - Vercel = deploying. (Free) - Namecheap = domain. ($12/yr) - Stripe = payments. (2.9%/transaction) - GitHub = version control. (Free) - Resend = emails. (Free) - = AI job Agent （Free） - Clerk = auth. (Free) - Cloudflare = DNS. (Free) - PostHog = analytics. (Free) - Sentry = error tracking. (Free) - Upstash = Redis. (Free) - Pinecone = vector DB. (Free) Total monthly cost to run a startup: ~$20 There has never been a cheaper time to build.

“AREX: Towards a Recursively Self-Improving Agent for Deep Research” Deep research agents often fail not because they need more search, but because they don’t know which parts of an answer are already verified and which constraints are still unresolved. This paper turns research into a recursive loop: answer, verify constraint-by-constraint, preserve evidence, then refine only the weak parts. AREX adds learned context updates and step-aware RL, achieving extremely strong results with 4B and 122B-A10B models across various agentic benchmarks.

Inkling-small is out today! With SGLang, you can get 648 tok/s decode with DSpark (simulated acc len=4) and 288 tok/s w/o DSpark, under the same setup (8x @NVIDIAAI B200, TP 8, NVFP4, bs=1). What makes this model different is the size. 276B total with 12B active is a sweet spot for RL, and both LoRA and full-parameter training become well within reach. Miles is ready and verified for multimodal RL on Inkling-small, so you can turn your multimodal data into real capability gains. At ~1/4 the size, Inkling-small matches the bigger version in capability and even wins on some benchmarks. Run Inkling-small with SGLang, and customize it with Miles.

"Metis: Memory Foundation Model" Most AI agents still use memory as an external RAG-style module, so the model retrieves old text instead of actually remembering. This paper makes memory native to the Transformer. So past interactions are compressed into dynamic layer states and read through memory attention during normal forward passes. The model weights stay frozen at inference, but its memory state updates without gradients, giving the model persistent memory inside the backbone. Still early and lossy, but this is yet another paper with a big step toward agents that remember natively instead of outsourcing memory to a database.

🔥 𝐍𝐞𝐰 𝐛𝐥𝐨𝐠: 𝐓𝐨𝐰𝐚𝐫𝐝𝐬 𝐋𝐨𝐨𝐩𝐞𝐝 𝐌𝐨𝐝𝐞𝐥𝐬 𝐃𝐨𝐧𝐞 𝐑𝐢𝐠𝐡𝐭 — 𝐏𝐚𝐫𝐭 𝐈 Looped models reuse the same weights across depth, promising a better compute–parameter trade-off, especially for reasoning. 𝐁𝐮𝐭 𝟏) 𝐝𝐨 𝐭𝐡𝐞 𝐠𝐚𝐢𝐧𝐬 𝐬𝐮𝐫𝐯𝐢𝐯𝐞 𝐰𝐡𝐞𝐧 𝐛𝐨𝐭𝐡 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐚𝐧𝐝 𝐢𝐧𝐟𝐞𝐫𝐞𝐧𝐜𝐞 𝐅𝐋𝐎𝐏𝐬 𝐚𝐫𝐞 𝐦𝐚𝐭𝐜𝐡𝐞𝐝? 𝟐) 𝐀𝐧𝐝 𝐰𝐡𝐢𝐜𝐡 𝐚𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐚𝐥 𝐜𝐡𝐨𝐢𝐜𝐞𝐬 𝐚𝐜𝐭𝐮𝐚𝐥𝐥𝐲 𝐦𝐚𝐭𝐭𝐞𝐫? We run 𝐚𝐩𝐩𝐥𝐞𝐬-𝐭𝐨-𝐚𝐩𝐩𝐥𝐞𝐬 ablations spanning Ouro to Huginn. Huginn performs better overall, with the largest gains coming from the loop-in-the-middle (sandwich) design and input injection, though they provide different benefits. Trained on 𝟓𝟎𝟎𝐁 tokens, an 𝟖𝐁-𝐀𝟎.𝟖𝐁 Huginn MoE approaches or surpasses a 𝟑𝟐𝐁-𝐀𝟑.𝟐𝐁 feedforward MoE on several reasoning benchmarks, including GSM8K (83.6% vs. 80.8%), while using 𝟕𝟓% 𝐟𝐞𝐰𝐞𝐫 resident parameters under 𝐦𝐚𝐭𝐜𝐡𝐞𝐝 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐚𝐧𝐝 𝐢𝐧𝐟𝐞𝐫𝐞𝐧𝐜𝐞 FLOPs. More details and the blog link in the thread ↓

We started this moonshot project a year ago. Now we are excited to share our progress on robot learning from egocentric human data 🕶️🤲 Key idea: Egocentric human data is robot data in disguise. By bridging the kinematic, visual, and distributional gap, we can directly leverage human data to scale up imitation learning. My thoughts: 1. Human data is robot data: We perceive with our eyes and act with our body and hands. If we can learn across robot embodiment, why not humans? 2. Data collection needs to be passive: Just as the Internet evolved into an unintentional data repository for AI, we envision systems that effortlessly capture embodied experiences from human activities, without humans’ conscious participation. 3. The data capturing technology is ready: Project Aria glasses by @RealityLabs @meta_aria capture all the information we need to turn human data into robot data: fisheye camera, 3D hand tracks, SLAM, … We will see more ubiquitous devices entering the consumer market. 4. The next generation foundation model will learn from embodied human data: The human sensorimotor experience is largely missing from today’s AI training data. Next-gen foundation models will understand humans from a human’s perspective, driven by large-scale embodied human data. See @simar_kareer's thread for more details about this project.

Introducing EgoVerse: an ecosystem for robot learning from egocentric human data. Built and tested by 4 research labs + 3 industry partners, EgoVerse enables both science and scaling 1300+ hrs, 240 scenes, 2000+ tasks, and growing Dataset design, findings, and ecosystem 🧵

Introducing Hallmark! An open source design skill to make beautiful UIs and landing pages by default. Works in Claude Code, Cursor, and Codex. npx skills add nutlope/hallmark

Introducing Supabase Evals. Our benchmark for how well AI coding agents build with Supabase. We run agents like Claude Code, Codex, and Open Code against real tasks and score what they do.

Antigravity is being used for research eval analysis. Instead of manually writing scripts and parsing tables, researchers can now analyze evals in minutes using natural language enabled by skills. Antigravity performs the quantitative analysis, proposes hypotheses, and spins up dynamic subagents in parallel to isolate the failure modes.

在 AGENTS/CLAUDE MD 里面加一句： Every implementation task that changes files MUST end with a git commit before the final response. - Inspect `git status` before editing and treat pre-existing or concurrent changes as user-owned. - Review the final diff and run proportionate verification before committing. - Stage only files or hunks that belong to the current task. Never bundle unrelated changes unless the user explicitly asks. - Use a concise descriptive commit message on `main`, report the commit hash, and do not push, amend, or rewrite history unless asked. - Read-only tasks and tasks with no file changes do not create empty commits.