What’s new, what matters, and what to build next.
SafetyPulse proposes monitoring a model’s refusal circuit throughout fine-tuning, then using an audited low-rank rollback when benign adaptation weakens safety behavior.
Article·6 minMOSS-Live frames real-time assistants as two coordinated systems: a conversational foreground that keeps listening and a background that completes longer tasks. Here is how to evaluate that design beyond a smooth demo.
Article·6 minAi2’s ToyBlocks framework uses a deliberately constrained tabletop world to generate demonstrations, separate planning from control, and expose where robot-learning systems fail to generalize.
Article·6 minA new harness-adapter recipe trains compact LoRA policies inside three coding-agent runtimes, combines verifiable tasks with tool-call safeguards, and exposes the practical tradeoffs behind more efficient local agents.
Article·6 minThe Open ASR Leaderboard has added private conversational speech for Egyptian and Nigerian English. The update is small in hours but important in method: it makes accent coverage visible while limiting optimization against public test audio.
Article·5 minSuperCLUE’s November table has different leaders for coding, reasoning, hallucination control, instruction following, and long context. The useful lesson is not a universal winner, but a better way to build workload-specific model trials.
Article·6 minLeRobot 0.7 links streamed datasets, distributed training, remote inference, and language-guided rollouts into one robotics workflow. Here is how to evaluate the new stack without confusing integration breadth with proven performance.
Article·5 minNVIDIA’s four Cosmos 3 Nano checkpoints cover driving, simulator rendering, and bimanual robotics. The useful question is not which demo looks best, but which model boundary matches the decision loop you need to evaluate.
Article·5 minPhonon's small, multilingual text-to-speech models shift the deployment question from raw model quality to a broader balance of latency, privacy, licensing, hardware coverage, and voice control.
Article·6 minDharma Field Extractor turns document images into typed JSON, illustrating how smaller specialized models can simplify extraction pipelines while leaving validation, abstention, and operational testing firmly in scope.
Article·6 minExplore releases and guides from 2024 onward. Dates refer to the original announcements; each article also shows when our coverage was published. How we cover the ecosystem ↗
