Daily Research Brief 2026-08-09

📊 Token usage: estimated from retrieval and writing scale.

Covers the latest AI research, open source and industry moves, updated daily.

Editor’s Note

The item worth pausing on today is one that seems unrelated to model capability: AWS is experiencing a CPU shortage internally. Engineer wait times for a dev machine went from hours to days, spot instances are scarce, and Intel’s own data shows the CPU:GPU ratio for AI inference shifting from 1:4 to something far more CPU-heavy in three months.

1. Latest arXiv Papers

1. Recursive Synthesis for Long-Horizon Terminal Tasks (RST) — https://arxiv.org/abs/2608.05466

2. ABSeeker: Answer-Backtracked Credit Assignment for Deep Search Agents — https://arxiv.org/abs/2608.05102

3. Voice Memory for Agentic Speech Recognition — https://arxiv.org/abs/2607.26410

4. OSReward: Cross-Platform Reward Models for Computer-Use Agents — https://arxiv.org/abs/2607.28609

5. GST-Bench: Global Spatial Awareness from Video — https://arxiv.org/abs/2608.05747

6. ProVisE: Generative Pixel-Space Evaluation of Spatial Cognition — https://arxiv.org/abs/2607.21072

7. WorldClaw: Agentic 3D Open-World Generation at Scale — https://arxiv.org/abs/2608.05248

8. Interpretable MEG Decoding of Perceived Speech — https://arxiv.org/abs/2608.01481