Research

Agentic AI, read daily · organised into threads

Looped flows enable deeper reasoning by chaining denoising steps

Suleymanzade et al. propose looped flows, a method that trains recurrent neural networks for many-step inference using local denoising losses. This overcomes the difficulty of training early updates to support later ones when backpropagation is truncated. The authors report state-of-the-art accuracy among looped models on six reasoning benchmarks, including 58.8% on ARC-AGI-1 and 12.2% on ARC-AGI-2.

Ayhan Suleymanzade, Chanhyuk Lee, Floor Eijkelboom +3arXiv →

AI & ML

Diffusion models generate high-resolution elevation maps from low-resolution inputs guided by optical imagery

The authors propose a guided super-resolution approach for digital surface models (DSMs) using denoising diffusion, improving coarse 5 m DSMs to 0.5 m resolution by leveraging high-resolution optical imagery. Experiments on several Central European cities show the method produces DSMs with crisper building outlines and more detailed roof structures than conventional interpolation or filtering.

2 days ago · arXiv

Memory & Planning

Agent-side memory steers stateless robots through long manipulation tasks

Hu et al. test whether long-horizon robot manipulation requires memory inside the action policy itself. On LIBERO-Mem, their architecture places all interaction history in a multimodal agent and achieves 76.3% average completion with an episodically stateless action model, versus 14.8% for the strongest reported baseline.

2 days ago · arXiv

Safety & Evals

Hidden prompt revisions inject cultural stereotypes into generated images

The authors introduce WORLDVIEW, a benchmark of 8,960 prompts spanning 15 languages and 31 language–context pairings, to examine how text-to-image systems rewrite requests before generation. Across DALL-E-3, Imagen-4, and GPT-Image-1.5, revisions marked non-Western and non-Anglophone settings more heavily than the United States and repeatedly reduced them to narrow, stereotypical vocabularies.

2 days ago · arXiv

Multi-Agent Systems

Backward Bayesian reasoning helps LLM agents resolve conflicting diagnoses

Chen et al. construct a reverse, label-free posterior from an explicit likelihood and compare it with agents’ conventional forward predictions. Across five LLM backbones, combining the two reasoning paths performed better than forward-only aggregation, particularly when agents disagreed.

2 days ago · arXiv

Agent Harnesses

Compressing Shared Sandbox Memory Makes High-Fanout Agents More Efficient

Li et al. designed AgentZip to reduce the memory consumed when agent tasks spawn many related sandbox sessions. Across LLM training and inference workloads, the system reduced sandbox-owned memory by up to 8.7-fold while limiting the slowdown from aggressive compression to 1.40-fold.

2 days ago · arXiv

Safety & Evals

Deep research agents struggle with long, multimodal evidence chains

The authors introduce a benchmark whose questions require an average of 12.1 intermediate conclusions and include non-text evidence that affects the reasoning process. The strongest evaluated system reached 43.1% Overall Accuracy but only 34.3% Strict Accuracy, exposing failures that final-answer scoring can obscure.

2 days ago · arXiv

All research articles →

Hot Thread

29 papers

Memory Management for Agents

papers per week · 8 weeks

Recent work establishes that agent memory can be effectively structured at multiple levels—kernel-managed personalization (Rutgers), hierarchical query-dependent summarization (Chinese Academy of Sciences), and game-theoretic filtering of harmful memories—while compression techniques (HKUST AgentZip) and training-free forgetting mechanisms (Beihang) improve efficiency under practical budgets. The latest papers extend these ideas to long-horizon robotics (KU Leuven), streaming video latent tokens (NUS, CUHK, Nanjing, Ant Group), and enterprise data staleness (ChurnBench), showing that memory design must be domain-specific and refresh-aware. Open questions remain about maintaining memory transferability across model upgrades—structured knowledge graphs appear robust but compressed notes do not—and about how to autonomously resolve conflicting or outdated memories without degrading task performance.

All threads →

The Field

Looped flows enable deeper reasoning by chaining denoising steps

Diffusion models generate high-resolution elevation maps from low-resolution inputs guided by optical imagery

Agent-side memory steers stateless robots through long manipulation tasks

Hidden prompt revisions inject cultural stereotypes into generated images

Backward Bayesian reasoning helps LLM agents resolve conflicting diagnoses

Compressing Shared Sandbox Memory Makes High-Fanout Agents More Efficient

Deep research agents struggle with long, multimodal evidence chains

The Field

Agent Harnesses

Compressing Shared Sandbox Memory Makes High-Fanout Agents More Efficient

Embedded software agents need interaction contracts and continuous assurance

VikingRAG Cuts Token Use for Retrieval Over Structured Documents

Similarity-aware context windows improve model routing across multi-turn conversations

Tools & MCP

Closed-loop synthesis efficiently trains language models to use external tools

Structured software interfaces outperform screenshot-and-click control for AI agents

Adaptive tool use improves video research agents’ accuracy and efficiency

Joint training helps small language models create and use tools

Multi-Agent Systems

Backward Bayesian reasoning helps LLM agents resolve conflicting diagnoses

Task-specific hierarchies improve coordination in large embodied AI teams

LLM teammates retain performance after swaps but coordinate less efficiently

Shared infrastructure spread both cheating and resistance through an AI swarm

Creative Agents

Identity preservation remains a distinct challenge for generative image models

Image-only pretraining helps build a strong open image generator

Coding agents combine generated imagery with editable web-based visual layouts

Language enables precise character and camera control in video worlds

Memory & Planning

Agent-side memory steers stateless robots through long manipulation tasks

Kernel-managed memory personalizes multiple AI agents with shorter prompts

Agents can retain old facts without using them indiscriminately

Hierarchical memory trees improve long-context reasoning without model training

Safety & Evals

Hidden prompt revisions inject cultural stereotypes into generated images

Deep research agents struggle with long, multimodal evidence chains

Refresh schedules, not cache age, determine agent answer staleness

Cross-device workflows expose major weaknesses in today’s GUI agents

RL for Agents

Enumerating Tool Choices Beats Sampling Them in Genomic Reasoning

Synthetic rewards train agents to diagnose simulated advertising anomalies

Adaptive rollout trees broaden language models’ mathematical reasoning coverage

Factorized reinforcement learning improves vision-language models’ spatial reasoning

AI & ML

Looped flows enable deeper reasoning by chaining denoising steps

Diffusion models generate high-resolution elevation maps from low-resolution inputs guided by optical imagery

Sparse mixture-of-experts models overfit repeated training data faster

Chain-of-thought reasoning traces are legible but not interpretable