Papers arxiv:2605.09820 Dystruct: Dynamically Structured Diffusion Language Model Decoding via Bayesian Inference Published on May 10 Submitted by Kevin Zhai on May 12 Authors: , Kevin Zhai , , Abstract A Bayesian structured decoding framework enables flexible-length generation in diffusion languag… …
Papers arxiv:2605.11518 AutoLLMResearch: Training Research Agents for Automating LLM Experiment Configuration -- Learning from Cheap, Optimizing Expensive Published on May 12 Submitted by taicheng guo on May 13 University of Notre Dame Authors: Taicheng Guo , , , Abstract An agentic framework calle… …
… This preserves tool-call fidelity, gives cross-round consistency, and locks in first-turn protection even if the on-device LLM is later compromised. 📊 Results Qwen3-4B + Gemini 3 Flash 📈 +15-36% accuracy and 2-6× lower leakage vs SOTA device-cloud baselines on $\tau^2$-Bench Airline/Retail and GAIA… …
… It's collaborative intelligence , where multiple independent LLMs across device and cloud exchange natural language or structured messages at the task level. 🧭 Two axes, one taxonomy We organize the design space along two composable dimensions: 📡 Vertical device-cloud collaboration : heuristic rout… …
… These results suggest that failed trajectories can provide structured repair supervision for safer self-evolving agents. …
… Across in-domain and out-of-domain evaluation, multi-token DiffRetriever substantially improves over single-token on every diffusion backbone we test, while autoregressive multi-token is flat or negative and pays a latency cost that scales with K where diffusion does not. …
… Existing methods typically rely either on costly external supervision from humans or strong teacher models, or on self-play training with static-difficulty instructions that cannot evolve as the model's capabilities improve. …
… All shaping is training-only; inference cost is unchanged. …
… Seeing Open Deep Research as an open-source, transparent alternative alongside a guide that helps with realistic cost estimation gives a really balanced view: you show how to build powerful AI tools and how to plan for their usage sustainably."