Daily Research Brief 2026-05-25

📊 Token usage: estimated from retrieval and writing scale.

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

Editor’s Note

Cohere open-sourced Command A+ (218B, Apache 2.0); Google’s Gemini for Science hit two Nature covers with ERA and Co-Scientist; arXiv banned AI-hallucinated citations; and OpenAI’s reasoning model cracked the 80-year-old Erdős unit-distance conjecture. Chinese universities swept the top 4 in global AI paper output.

1. Latest arXiv Papers

1. StableEdit: Lifelong Normalization Enables Million-Scale Edits Without Collapse — https://arxiv.org/abs/2605.11836

2. LingBot-VA: A Causal World Model for Robot Simulation — RSS 2026 — https://arxiv.org/abs/2601.21998

3. ERA Framework: Intelligent Patent Value Assessment, Breaking the Tech-Legal Double Black Box

4. Cohere Open-Sources Command A+ (218B Parameters, Apache 2.0)

5. Google’s Gemini for Science Lands Two Nature Covers — ERA Assistant & Co-Scientist

6. arXiv Bans Fake Citations: AI-Hallucinated References Earn One-Year Posting Ban

7. Cambridge Study: AI Still Can’t Grade University Essays — Form-Over-Content Bias

8. Google’s Science AI Roadmap: From Specialized Tools to Autonomous Scientists

9. Chinese Universities Sweep Top 4 in Global AI Paper Output

10. OpenAI’s General Reasoning Model Cracks the 80-Year-Old Erdős Unit-Distance Conjecture