Researchers from Peking University and Google Cloud AI have introduced PaperBanana, a system of five AI agents that collaboratively generate publication-ready diagrams and charts for academic papers.
The system links agents responsible for retrieval, planning, styling, rendering, and critique, mimicking the human designer’s process when creating diagrams. On a benchmark of 292 NeurIPS methodological diagrams, PaperBanana exceeded standards by 37% in conciseness and nearly 13% in readability. It can also enhance hand-drawn human figures, winning aesthetic comparisons 56% of the time.
Tools like PaperBanana and OpenAI’s Prism demonstrate how the process of preparing diagrams and illustrations for scientific work is accelerating, freeing researchers to focus on ideas instead of technical barriers to content creation.
In Brief: Tech World Highlights
- Daniela Amodei of Anthropic stated in a CNBC interview that “exponential growth continues until it stops… and so far it has continued every year” in AI development.
- Boston Dynamics and Google DeepMind announced a partnership to integrate the Gemini Robotics AI model into Atlas humanoid robots.
- OpenAI researcher Jerry Tworek revealed that he is leaving the company after seven years, during which he contributed to the first programming systems and led the AI reasoning team.
- Claude Code creator Boris Cherny released a guide on using this agent tool for coding, including managing up to 15 parallel sessions simultaneously.
- OpenAI Chief Product Officer Fidji Simo outlined the 2026 product roadmap in a new blog, stating the goal of turning ChatGPT into a proactive “personal super-assistant.”
AI Trending Tools:
- ChatGPT Health: OpenAI’s new private experience for healthcare conversations.
- Manus Minutes: Converts live conversations into structured notes.
- Copilot Checkout: Enables completing purchases directly within Microsoft Copilot.
