#论文/研究

Pen spinning on WUJI Hand 2, from simulation training to real-world execution. We use motion-reference tracking and reinforcement learning to bring a simulation-trained policy to the physical hand. Code and setup guides are open source for developers and researchers to build on. Code: http://github.com/wuji-technology/wuji-mjlab

🤖 𝗘𝘃𝗲𝗿𝘆 𝗿𝗼𝗯𝗼𝘁 𝘁𝗮𝘀𝗸 𝗳𝗮𝗶𝗹𝘂𝗿𝗲 𝗰𝗮𝗻 𝗯𝗲𝗰𝗼𝗺𝗲 𝗮 𝗹𝗲𝘀𝘀𝗼𝗻 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝗶𝘁𝗲𝗿𝗮𝘁𝗶𝗼𝗻. 𝗦𝗲𝗻𝘀𝗲𝗡𝗼𝘃𝗮-𝗥𝗼𝗯𝗼𝗥𝗦𝗜 uses 𝗥𝗦𝗜 to evolve 𝗮𝗴𝗲𝗻𝘁 𝗵𝗮𝗿𝗻𝗲𝘀𝘀𝗲𝘀 through physical task feedback, without updating model parameters. ▶ 𝗥𝗼𝗯𝗼𝗗𝗼𝗷𝗼: 56.83 avg. score, 50.83% success rate, +𝟮𝟳.𝟴𝟲 𝗽𝗼𝗶𝗻𝘁𝘀 𝘃𝘀. 𝘁𝗵𝗲 𝗚𝗣𝗧-𝟲 𝗔𝘀𝘁𝗿𝗮 𝗯𝗮𝘀𝗲𝗹𝗶𝗻𝗲 (𝟮𝟴.𝟵𝟳) ▶ 𝗥𝗦𝗜 𝗹𝗼𝗼𝗽: Execute → Diagnose → Improve → Validate → Inherit ▶ 𝗘𝘅𝗽𝗹𝗼𝗿𝗲𝗱 𝗮𝗻𝗱 𝘃𝗮𝗹𝗶𝗱𝗮𝘁𝗲𝗱: Multi-point EEF prediction and Planning + Feedback Subagents This achievement reflects the collective efforts of our joint team. RoboRSI will be 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲𝗱 𝘀𝗼𝗼𝗻 at https://github.com/OpenSenseNova/SenseNova-RoboRSI. A comprehensive 𝘁𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗿𝗲𝗽𝗼𝗿𝘁 detailing our methods and findings is also on the way. Stay tuned!

Pangram, the AI-text detector, missed 79.8% of scientific abstracts rewritten by Meta's Muse-Glimmer, while flagging just 1 of 5,000 human abstracts. In a new paper from Tokyo Metropolitan University, reseaerchers find the share of AI-rewritten abstracts that Pangram misses depends strongly on the LLM version Shows that it caught 93.5% of GPT-5 rewrites but missed 79.8% from another new model. Its miss rate depended mostly on which model did the rewriting.

I've been thinking a lot about making robot learning scalable. To me, that means learning on the job, and using the human help still needed efficiently. In Mulligan ⛳️, we focus each data-collection round on the initial states where the robot fails. This happens in deployment, with the robot doing the task and a person stepping in when needed. On 3 real tasks and 2,550 blind evals, it improves on uniform collection at the same budget.

New paper. Previous: Models transfer a love of owls through sequences of numbers (Subliminal Learning). New: We show models can transfer more complex traits: novel skills, agentic hacking, backdoors. Here, a backdoor transfers with neither trigger nor behavior in the data!

New Research: We are releasing ExplorationBench, a benchmark for measuring how AI systems explore. Scientific discovery begins where known problems end: a system has to frame hypotheses, design experiments, and learn from the results. Evaluating this is hard. Genuinely new answers cannot be checked quickly, and in familiar domains a model can simply recall what it has seen. Addressing this challenge, researchers from Tencent Hy, Fudan University, and Tsinghua University built verifiable Alien Worlds. Their rules are executable, so every answer is checked exactly, and they conflict with familiar knowledge, so recall alone cannot solve the tasks. 🔹 Two sandboxes: AlienCode (31 hidden rule changes, 70 tasks) and AlienLogic (24 patched inference rules, 70 theorems) 🔹 A flawed manual, four rounds of self-designed probes, and closed-book tests after every round 🔹 Every answer graded by an interpreter or a proof checker, with no LLM judge What we found across 10 frontier AI systems: 1️⃣ Getting feedback is more effective than thinking alone. No AlienCode run starts above 15.7%; after four rounds the best reaches 89.0%, while the same turns without feedback stay at 0.5–11.0%. 2️⃣ Designing the experiments matters. Replaying a system's own best probes gives it exactly the same evidence, yet in AlienCode 9 of 10 systems do worse than when they chose the probes themselves. 3️⃣ Knowing a rule is not using it. Even when every required rule is stated correctly, tasks are solved only 73.4% of the time. 4️⃣ One score hides a lot. The same system under the same budget ended anywhere from 5.7% to 79.0%, and rankings barely transfer between the two worlds. CL-bench asked whether models can learn from context. ExplorationBench asks whether they can discover the rules themselves. 📄 Paper: https://arxiv.org/abs/2609.30199 🌐 Website & leaderboard: https://explorationbench.com 📝 Blog: https://explorationbench.com/blog/ 💻 Code (coming soon): https://github.com/Tencent-Hunyuan/ExplorationBench