AI Briefing, April 9 · Issue #188
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
**Gemini Nano** is accelerating the deployment of lightweight, on-device AI applications—enabling real-time, personalized sticker generation and other interactive use cases. **Qwen3.6-Plus** has officially reached production readiness, delivering significant improvements in **latency and inference capability**. Meanwhile, **ALTK-Evolve**’s novel long-term memory mechanism for AI agents introduces a new paradigm for reliability in complex, multi-step tasks [4][15][5].
## 🚀 Highlights
- **DeepLearning.AI launches a free course on efficient LLM inference with SGLang** [0]: Covers **SGLang**, **KV caching**, and **RadixAttention**, offering a systematic deep dive into optimization techniques for LLMs and image generation models.
- **Google DeepMind demos Gemini Nano–powered personalized stickers** [4]: Showcased live at AI Engineer Europe—sticker generation runs entirely on-device using **Gemini Nano** + Google AI Studio.
- **ALTK-Evolve: On-the-job learning for AI agents** [5]: Introduces a **long-term memory subsystem** that distills interaction histories into reusable principles—boosting reliability across multi-step workflows.
- **A humanities professor at Wuhan University makes a bold AI pivot** [6]: Prof. Qiong Wang’s team launched *AiCharts*—a “white-box” solution tackling AI chart **hallucinations** and **non-editability** head-on.
- **Qwen3.6-Plus officially launches—production-ready** [15]: Alibaba announces the model has exited beta: **lower latency**, **stronger reasoning**, and full support for enterprise-scale deployment.
- **Moxt: An agent-native collaboration workspace & AI employee manager** [8]: A new AI-native platform enabling users to “hire” **AI employees**, automate tasks, and auto-archive information.
- **Rowboat: Open-source AI assistant for building personal knowledge graphs** [22]: A Y Combinator–backed project that continuously constructs a **context-aware knowledge graph** by connecting your work tools.
- **Genspark AI enforces a 5-hour session limit for Plus/Pro users** [11]: A new policy to ensure service stability and prevent abuse amid rapid growth.
## 🔗 Sources
[0] DeepLearning.AI Launches Free Course on Efficient Inference with SGLang — https://www.bestblogs.dev/status/2041901657114423338
[4] Google DeepMind Demonstrates Gemini Nano–Powered Personalized Stickers — https://www.bestblogs.dev/status/2041891745558827095
[5] ALTK-Evolve: On-the-Job Learning for AI Agents — https://www.bestblogs.dev/article/58f3e316
[6] A Humanities Professor at Wuhan University Makes a Bold AI Pivot: Project Volume Up 4000% in Six Months, Redefining Chart Productivity — https://www.bestblogs.dev/article/52b6eaac
[8] Moxt: Agent-Native Collaboration Workspace & AI Employee Management — https://www.bestblogs.dev/status/2041882523429695855
[11] Genspark AI Enforces 5-Hour Session Limit to Meet Growing Demand — https://www.bestblogs.dev/status/2041874874336223659
[15] Qwen3.6