Analyzes on-policy distillation for LLM fine-tuning, shows teacher token-level supervision is often noisy and not the main driver of gains, and introduces OPSA, a supervision-free, entropy-adaptive method that suppresses low-probability tokens to improve downstream accuracy.
AI Agent Papers2026
Turns each research paper into a training environment to generate verifiable research plans by synthesizing questions from goals/background and deriving evaluation criteria from methods/experiments. Key features: four-stage extraction that reduces criterion leakage to 3.7%, a two-stage rubric-centered training (self-distillation then GRPO), and the PaperGym-20k corpus with two held-out benchmarks.
Develops a vision-language foundation model for autonomous driving that unifies 3D BEV perception, visual question answering, and motion planning without changing the pretrained VLM architecture. Key elements include an external BEV perception head for 3D detection and occupancy, a Planning Expert using flow-matching for trajectory prediction, and a staged training recipe combining driving and general VLM data.
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
A TURBO multi-stage fine-tune of Qwen3.8‑27B that shortens internal “thinking” token blocks and raises ARC benchmarks (8‑bit ARC‑C ≈735, ARC‑E ≈882). It ships GGUF quants (regular and MTP, Neo‑Imatrix), vision support and 256k context for local multimodal inference on consumer GPUs.
Compresses conversational histories and long documents into short sequences of continuous soft memory tokens that a frozen decoder can read directly without text reconstruction. Uses a small reader-matched writer that trains only a tiny adapter, achieving 4–16× compression and much faster write/read latencies.
AI Model2026
NVFP4-quantized checkpoint of Qwen3.8-Flash-Next for GPU-optimized multimodal autoregressive inference — routed MoE experts in W4A4 NVFP4 while attention/ancillary layers remain BF16; ~2.7× smaller than the BF16 source and supports very long contexts.
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.
AI Agent Papers2026
Reconstructs executable terminal workspaces from recorded agent trajectories and synthesizes verifiable single- and multi-round coding tasks for agent training; it replays file operations, uses an LLM completion agent to fill missing files/dependencies, and verifies tasks with autogenerated test suites.
Converts a natural-language function specification into a reusable local neural function by using teacher models to synthesize examples and finetuning a small adapter for a compact interpreter. Achieves higher semantic accuracy (83.6% on FuzzyBench-Hard) at the cost of roughly one minute compile time; produces versionable PAW artifacts for local deployment.
Combines sparse verifier outcomes with dense privileged‑hindsight token scoring to learn an outcome‑calibrated, normalized distribution over complete responses for on‑policy self‑improvement. Key features: sign‑gated guidance (retain/reverse/disable per verifier advantage), profiled trajectory balance with one log‑partition per rollout group, and explicit correction against false‑positive self‑guidance.