AI Briefing, March 31 — Issue #161
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
China’s AI capabilities are advancing on two fronts: foundational models and real-world industrial adoption. **Qwen3.5-Omni** outperforms **Gemini-3.1 Pro** across multimodal benchmarks, while **PaddleOCR** has become the most-starred OCR project on GitHub globally. Meanwhile, **InCoder-32B** fills a critical gap in industrial-scale code LLMs for chip design—and Insilico Medicine’s **$2.75 billion** AI-driven drug discovery partnership with Eli Lilly signals that AI biotech has officially entered its commercialization phase [2][3][15][18].
## 🚀 Top Updates
- **Alibaba launches Qwen3.5-Omni—multimodal performance surpasses Gemini-3.1 Pro** [2]: Achieves state-of-the-art results on 215 audio/video understanding tasks, with strong cost-performance advantages.
- **Baidu’s PaddleOCR tops GitHub’s global OCR repository rankings (73.3k+ stars)** [3]: A lightweight, high-accuracy solution powered by ERNIE large models—marking China’s first historic breakthrough in an AI foundational technology category via open source.
- **Beihang University open-sources InCoder-32B, an industrial-grade code LLM** [15]: A 32-billion-parameter model trained on 2.5 million real-world simulation datasets—optimized for chip design, GPU optimization, and embedded systems.
- **Insilico Medicine and Eli Lilly ink a $2.75 billion drug R&D deal** [18]: Triggers U.S. Hart-Scott-Rodino antitrust review—underscoring both the scarcity and commercial maturity of AI-powered drug discovery platforms.
- **Microsoft introduces Copilot Critique—a multi-model research system** [4]: Uses a “generate + evaluate” dual-AI workflow to significantly improve accuracy and trustworthiness of deep research within Microsoft 365.
- **Volisphere launches the first auction-based ad platform built for AI agents** [16]: Organizes ads around reusable “Skills,” defining a four-layer evolution framework for advertising in the AI-native era.
- **Microsoft Research releases three interactive AI implementation handbooks (Paza/Vibhasha/Atlas)** [20]: Focus on cross-lingual and cross-cultural deployment—advancing inclusive, production-ready AI engineering.
- **Claude expands desktop task execution capabilities** [22]: Adds application navigation and local file/system operations—extending AI’s practical utility as a true “digital employee.”
## 🔗 Sources
[1] AI Should Be a Good Citizen, Not Just a Good Assistant — LessWrong — https://www.bestblogs.dev/article/f8f9a887
[2] Alibaba Launches Qwen3.5-Omni—Multimodal Performance Surpasses Gemini-3.1 Pro — https://www.bestblogs.dev/article/cc80f169
[3] The New Global OCR Champion Is Chinese Open Source! GitHub Hits 73,300+ Stars — https://www.bestblogs.dev/article/7c6d24d4
[4] Microsoft Launches Copilot Critique to Strengthen Deep Research Capabilities — https://www.bestblogs.dev/status/2038615654614348263
[15] InCoder-32B Open-Sourced: A 32-Billion-Parameter Industrial Code LLM Filling Gaps in Chip Design and Embedded Systems — https://www.bestblogs.dev/article/05dab9a6
[16] The First Auction-Based Ad Platform for AI Agents: Volisphere Believes the Next Era Will Birth a New AppLovin — https://www.bestblogs.dev/article/8763ecb0
[18] Behind the Nearly $2 Billion Deal: Insilico Medicine’s Strategic Partnership with Eli Lilly — https://www.bestblogs.dev/article/94dadf9