• Self-HPO Raises Qwen3.5-35B-A3B's Average Resolve Rate Across Three Coding Benchmarks From 57.8% to 64.8%. It alternates between updating the harness and policy. Switching to a stronger external harness proposer delivers a further boost.

• Looped Models Need the Right Topology, Input Injection, and State Design. With matched training and inference compute, the full Huginn-style design uses 75% fewer resident parameters than a 112-layer feed-forward MoE. It performs better on DROP and GSM8K while tying on MATH500.

• Synthetic Attribute-Binding Text Can Improve Visual Retrieval. In this study, fine-tuning an MLLM-based embedder exclusively on such text improved cross-modal and image-to-image retrieval while preserving general multimodal embedding performance. Open-world generalization still needs testing.

Also Notable

• Human Posters Become Verifiable HTML/CSS Templates — RAP converts more than 25,000 human-designed paper posters into reusable slots and verifies outputs through rendering comparisons. Its held-out test set pairs 150 papers with manually designed posters..
• Specialized Critics Cut 3D Scene Trial and Error — Separate design and critique experience trains a scene agent. A dedicated critic then supports both reinforcement learning and test-time correction. Scene-OTA-35K and Scene-OTC-15K provide design and quality-critique supervision..
• Lean4 Verifies Rewards at Every Workflow Layer — The framework verifies tool calls, workflow stages, and final results with Lean4. Its stage-normalized GRPO trains LAR-9B and LAR-27B. The paper reports that LAR-27B reaches 77.3% on TerminalBench-2.1 and beats leading models of similar size..
• One Interface Unifies Video Editing References — A single reference-bank interface accepts variable numbers of text, image, and video inputs. It brings camera motion, subjects, style, and effects into one editing framework. R²-RoPE helps the model generalize to more references..
• State Propagation Breaks Text World Models — When actions affect dependent objects, average task success across 20 models falls to 5.1%. GlobeBench contains 1,632 tasks across five domains and 19 simulated skills. No model solves 631 of them..
• Harmless Self-Distillation Can Still Deepen Misalignment — Misalignment can worsen through recursive self-distillation even when training prompts and model outputs are harmless. The effect appears in both the Qwen3 and Phi-4 families. Self-generated data loops need their own safety evaluations..
• Deterministic Checks Protect Correct Formal Specifications — DRAGO diagnoses executable but semantically wrong specifications. It accepts revisions only when their error weight falls. This cuts damage to correct specifications from 11.8% to 1.3%. Across five benchmarks, answer accuracy beats zero-shot formalization by 12.9 percentage points..
• Experience Cuts Coding Agent Token Costs — Cross-task experience supplies micro-skills, removes irrelevant tools, and stops failed loops without modifying the black-box agent. Average token costs fall by 18.4% to 30.9% while success rates stay similar across Claude Code, Codex, and GitHub Copilot CLI..
• VISTA Tests Goal-Aware Egocentric Assistance — Vision-language models can often describe first-person scenes but struggle to help users achieve their goals. VISTA tests that gap with six synchronized sensor inputs, about 1,000 egocentric recordings, and zero-shot evaluations of 14 representative models..
• Answer Accuracy Hides Weak Audiovisual Grounding — Models with similar answer scores can differ sharply in whether annotated spatial, temporal, and audio evidence supports their explanations. ANCHOR contains 2,319 videos and 8,272 question-answer pairs. It separately measures correctness, hallucination, logical drift, and concision..
• Monitor Transfer Treats Model Differences as Symmetries — LEMON models attention-head permutations and depth differences as symmetries. Lightweight monitors can then transfer to new LLMs without fresh labels. Across 16 LLMs and six datasets, ROC-AUC beats the strongest baseline by as many as 14 points..

Read the full edition →