What's new
The 30 most-recently-filed papers and most-recently-updated concept pages.
Papers
- tweet Disabling thinking and exposing deep_think as a tool leaks the internal CoT reasoning format
- blog TwIL-8B — 8.8B Reasoning Model Post-Trained from Granite 4.2 with SFT + GRPO/DAPO RL
- arxiv Rethinking Activation Functions: Reducing Bias Can Improve Learning
- arxiv Beyond Validation Loss: Coverage and Reliability Dynamics in Robot Policy Pretraining
- arxiv Human-to-Robot Transfer via Bridging Representation and Action Space in VLAs
- arxiv Reinforcement Learning Enables Scaling Generalist Robot Policy Improvement
- arxiv Fast and Accurate Hand Estimation in Pixels, Interaction and World Space
- arxiv SLA: Scaling Spatial-Language-Action Model via Continual Pre-training
- blog J++ Lens: Jacobian Filtering enables more faithful workspace readouts
- blog cuRoboV2-Fast: Fast Torque-Limited Collision-Free Motion Planning
- blog Introducing Inkling 2 Small — Thinking Machines' 594B/17B MoE with unified RL at 4e23 FLOPs
- arxiv GameVerse: A Minute-Scale Gameplay Dataset for Long-Horizon Interactive World Modeling
- arxiv Multi-View Latent MoE: Letting Every Router Shape Expert Computation
- arxiv ODE Policy: Robot Learning via Action Dynamics, Not Actions
- arxiv DreamFree: Future Representation Distillation for Imagination-free World Action Models
- blog ERNIE 5.5 Flash Base
- blog The Ultimate MoE Training Handbook: training Mixture-of-Experts models with Megatron-LM, setting by setting
- blog VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
- blog Ripple-65B Preview — Open-Weight Dense 65B Co-Designed with Ricursive's Chip Platform
- tweet Perceptron Egocentric 1.0 — self-service API with SOTA hand tracking and EgoHome benchmark
- arxiv Inspect Robots: Evaluating the Capabilities and Safety of Embodied AI
- arxiv EgoHand: Monocular 3D Hand Reconstruction from Egocentric Video
- arxiv Robotic Data Inoculation
- arxiv Pixel-UMI: Learning Environment-Aware Whole-Body Manipulation from Sensorized Human Demonstrations
- arxiv Pixel: A Co-Designed Full-Stack Robot System for Learning General-Purpose Mobile Manipulation
- blog AIRoA-0: Long-Horizon, Precise, Cross-Embodiment Manipulation from Handheld Data
- blog GPU-accelerated Weight Teleportation
- blog EgoHand v1.0 — NVIDIA monocular 3D hand pose + MHR/SOMA foundation model for robotics
- tweet Embodied Analysis — GPT-6 Astra vs Claude Opus 5 on test-tube rack placement, scored on execution directness
- tweet Pranay Gupta — direct weight manipulation of π0.5 beats fine-tuning on cloth folding (50% → 68% FT → 75% weight edit)
Concept pages
- concept Inference-Time Scaling
- concept LLM Inference Efficiency
- concept Open foundation-model releases
- concept Reasoning RL
- concept SFT-for-RL Readiness
- concept Agentic Software Engineering
- concept AI-for-AI Research
- concept Human-to-Robot Retargeting
- concept Hyperparameter scaling laws
- concept IO-Aware Kernel Design
Want a digest? It’s not currently posted to Slack. If that becomes useful, we can add the weekly trend digest covered in the original plan.