Daily Research Brief 2026-04-18

📊 Token usage: estimated from retrieval and writing scale.

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

Editor’s Note

Anthropic’s relationship with the Trump administration is in focus; the App Store is booming again, possibly thanks to AI; Sam Altman’s project World looks to scale; Schematik is ‘Cursor for Hardware’ with Anthropic backing; OpenAI executive Kevin Weil is leaving.

1. Latest arXiv Papers

1. Bidirectional Cross-Modal Prompting for Event-Centric Video QA — https://arxiv.org/abs/2604.15312

2. LeapAlign: Post-Training Flow Matching Models for Alignment — https://arxiv.org/abs/2604.15311

3. Generalization in LLM Problem Solving: The Case of Few-Shot Learning — https://arxiv.org/abs/2604.15306

4. Diagnosing LLM Judge Reliability: Conformal Prediction Approaches — https://arxiv.org/abs/2604.15302

5. MM-WebAgent: A Hierarchical Multimodal Web Agent — https://arxiv.org/abs/2604.15309

6. How Do LLMs and VLMs Understand Viewpoint Rotation? — https://arxiv.org/abs/2604.15294

7. CoopEval: Benchmarking Cooperation-Sustaining Multi-Agent Systems — https://arxiv.org/abs/2604.15267

8. From Tokens to Steps: Verification-Aware Speculative Decoding — https://arxiv.org/abs/2604.15244