Embodied AI2026
Learns generalizable World Action Models for robotic manipulation by scaling causal egocentric video pretraining and grounding learned dynamics with heterogeneous robot trajectories. Key features: a three-stage curriculum (video pretraining, video-action mid-training with a unified action representation, and target-robot specialization) and a Slow–Fast dual-system for 30 Hz real-time action prediction.
Embodied AI2026
Wei Wang, Wenqiao Zhang +15Affiliation: College of Computer Science and Technology, Zhejiang University, Affiliation: Nanjing University of Aeronautics and Astronautics +4
Wraps vision–language–action policies into executable skills that are runtime-validated, executed as bounded low-level action chunks, outcome-verified, and logged as structured trajectories. A fixed skill interface enables swapping or adapting low-level VLA policies and provides component-level supervision for training and optional online adaptation.
AI Model2026
Sparse MoE causal LLM that uses Mixture-of-Value Attention (MoVA) to store 36B parameters while activating ~4B per token; supports a native 524,288-token context and is released with final checkpoints, training data, and training code under open license.
AI Dataset2026
Provides 10,000 agentic multi-turn coding and reasoning traces from Fable 5.1 with step-by-step chain-of-thought and tool-use, heavily deduplicated and filtered; totals ~500M tokens (1.97 GB). Suited for SFT, distillation, and training long-horizon reasoning and code-generation models.
AI Video Papers2026
Provides an end-to-end, reproducible foundation for camera‑controllable, long‑horizon video world models — converting 1.43M clips from 10 datasets into a unified canonical corpus and releasing data, pipelines, recipes, and weights. Introduces backbone‑native adaptation and a three‑stage training recipe to produce 5B–33B models that enable minute‑to‑hour rollouts after training on 5s sequences.
Introduces Declarative Attention (DA), a zero-shot protocol that has LMs declare which parts of long context to attend to during chain-of-thought, letting the runtime build dynamic attention masks and skip most KV-cache reads. Produces large token savings (up to ~52% on Gemma-4-31B) with modest accuracy loss.
AI Agent Papers2026
Presents two LLM-based search agents (Iris-mini and Iris-pro) trained by alternating supervised fine-tuning and reinforcement learning against live web search. Key features: web-graph-derived multi-hop tasks with entity abstraction, SFT–RL climbing, inference-time context management, and state-of-the-art open-source benchmark results.
Zixuan Fu, Bingxiang He +11Affiliation: University of Chinese Academy of Sciences, Affiliation: Tsinghua University +3
Studies on-policy distillation (OPD) at the data-minimal limit by training on a single query, measuring state coverage and alignment dynamics, and showing OPD is often data-overfed but algorithm-starved.
AI Model2026
A 4B causal language model post‑trained from Qwen3.5‑4B for agentic workloads — tool use, coding and instruction following — using a routing‑harness feedback loop aimed at iterative capability improvement; distributed as text-only safetensors with native 262,144-token context.
AI Dataset2026
Provides a bilingual Chinese–English corpus for LLM training covering pretraining, capability-oriented midtraining (16K–256K long contexts), and supervised fine-tuning. Includes ~4.2T pretrain tokens, ~600B midtrain tokens, and ~4.57M SFT samples; sources span web, PDFs/OCR, code, math, QA, and agentic trajectories under mixed upstream licenses.
AI Dataset2026
Provides ~86K verifiable-reward RL training samples across Math, Knowledge (STEM), Long-Context, and Code for post-training LLMs; each sample includes a ground-truth and verifier-friendly JSONL format for stable reward signals. Note: code tasks require an external sandbox to execute tests.
Investigates long-horizon memorization in language models by evaluating how complementary continual-learning mechanisms (data, function, weight anchors and low-rank allocation/LoRA) compose to reduce catastrophic forgetting across 100 sequential tasks; best composition raises final retention from 1.2% to 34.9%.