• AI KATANA
  • Posts
  • Gemini Robotics brings AI into the physical world

Gemini Robotics brings AI into the physical world

Also: Salesforce to invest $1B in Singapore to boost adoption of AI

Author

AI KATANA
March 12, 2025

Hello!

Today’s AI landscape is buzzing with groundbreaking advancements, from robotics to AI-driven scientific research. Google DeepMind has unveiled Gemini Robotics, integrating AI into real-world robotics with a focus on adaptability and safety, while Salesforce is committing $1 billion to Singapore to boost AI adoption, particularly in digital workforces. Meanwhile, Japan’s Sakana AI has made history with the first AI-generated, peer-reviewed scientific paper, signaling a new era of AI-driven research. On the enterprise front, ServiceNow’s Yokohama platform releaseemphasizes agentic AI to automate workflows, while Google’s Gemma 3 brings powerful open-source AI to a single GPU, offering an expansive 128K context window. In AI tools, Google’s Gemini 2.0 Flash impresses with its native multimodal image generation, enabling real-time edits and style transfers. Lastly, Bessemer Venture Partners has raised $350 million for its second India fund, focusing on AI-driven startups amid shifting investment trends.

Sliced just for you:

  • 🤖 Gemini Robotics brings AI into the physical world

  • 💰 Salesforce to invest $1B in Singapore to boost adoption of AI

  • 🇯🇵 Japan-based tech startup produces 1st AI-generated peer-reviewed scientific paper

  • 🚀 ServiceNow’s Yokohama platform release focuses on agentic AI

  • 🖥️ Google’s Gemma 3 is an open source, single-GPU AI with a 128K context window

🤖 Gemini Robotics brings AI into the physical world

Google DeepMind has introduced Gemini Robotics, a new AI model based on Gemini 2.0, designed to integrate AI into the physical world through robotics. This development includes two key models: Gemini Robotics, a vision-language-action system that enables robots to understand and execute tasks, and Gemini Robotics-ER, which enhances spatial reasoning for more precise robotic actions. These models aim to create more adaptable, interactive, and dexterous robots capable of responding to dynamic environments and performing complex tasks like object manipulation. DeepMind has partnered with Apptronik to develop humanoid robots and is working with other industry leaders to refine its technology. A strong emphasis is placed on safety, with built-in measures to ensure responsible deployment and a new dataset to evaluate AI-driven robotics safety. This marks a step toward more practical and human-aligned robotics applications in everyday settings.

Salesforce is committing $1 billion to Singapore over the next five years to accelerate AI adoption, primarily through its Agentforce platform. The initiative aims to address labor shortages by integrating AI-powered digital workforces into enterprises. This investment follows similar moves by the company in Saudi Arabia and Argentina, each receiving $500 million to expand AI and cloud services. Salesforce has been deeply engaged in Singapore since establishing its first overseas AI research hub there in 2019, serving major clients like Singapore Airlines, Grab, and FairPrice Group. A new partnership with Singapore Airlines will integrate Agentforce, Salesforce’s AI layer Einstein, and Data Cloud into the airline’s customer management system, with plans to co-develop AI solutions for the aviation sector. This expansion aligns with a broader trend of U.S. tech giants, including Amazon Web Services and Microsoft, significantly investing in Southeast Asia to enhance cloud and AI capabilities.

🇯🇵 Japan-based tech startup produces 1st AI-generated peer-reviewed scientific paper

Japanese tech startup Sakana AI has made history by producing the first AI-generated peer-reviewed scientific paper. Their AI system, AI Scientist-v2, successfully passed the rigorous peer-review process at the International Conference on Learning Representations (ICLR), marking a major milestone in AI-driven research. Developed as an upgraded version of AI Scientist, the model is designed to autonomously generate scientific research from concept to final report. Sakana AI collaborated with the University of British Columbia and the University of Oxford on this breakthrough. The company envisions future iterations of AI Scientist advancing scientific discovery at an unprecedented scale, potentially exceeding human research capabilities. This development signals a new era where AI plays a significant role in producing high-quality, peer-reviewed scientific contributions.

ServiceNow has launched the Yokohama release of its Now Platform, emphasizing agentic AI to enhance enterprise automation. This update builds on previous AI advancements by introducing preconfigured AI agents that streamline security operations, automate change management, and proactively detect and resolve network issues. A key addition is the AI Agent Studio, which enables developers to create and customize AI agents without extensive coding. The platform also integrates workflow automation through the AI Agent Orchestrator, improving enterprise-wide data connectivity. Other notable features include service observability, which consolidates monitoring tools for faster issue resolution, and self-service sales portals that allow customers to manage orders without human intervention. Industry analysts suggest ServiceNow is positioning itself as a strong alternative to Salesforce, particularly for companies looking to modernize legacy systems. This move marks a significant shift in enterprise AI, offering businesses a more autonomous, AI-driven approach to operations and service management.

Google has unveiled Gemma 3, an open-source AI model optimized for efficiency, capable of running on a single GPU or even a smartphone. Built on the Gemini 2.0 foundation, Gemma 3 boasts a massive 128,000-token context window and supports multimodal inputs, including text, images, and video. The model is available in different sizes, ranging from a lightweight 1-billion-parameter version to a powerful 27-billion-parameter variant. Google claims significant improvements over previous open-source models in areas like math, coding, and complex reasoning. Gemma 3 is accessible through Google AI Studio and platforms like Hugging Face, though its use is subject to licensing restrictions. To enhance safety, Google has integrated ShieldGemma 2, a filtering system that blocks harmful images. This release aligns with a growing trend toward efficient, locally deployable AI models, offering developers more control over AI applications while reducing computational costs.

🛠️ AI tools updates

Google has launched Gemini 2.0 Flash, a cutting-edge multimodal AI model that introduces native image generation within the same framework as text processing. Unlike traditional AI image models that rely on diffusion techniques linked to language models, Gemini 2.0 Flash generates images directly from text prompts, enabling greater accuracy and flexibility. Key features include text-image storytelling, real-time conversational image editing, world knowledge-based image generation, and improved text rendering within images. Early adopters have demonstrated its ability to generate and refine illustrations, edit images iteratively, and even modify specific elements like adding chocolate drizzle to a croissant in seconds. This release positions Google as a leader in AI-driven image generation, surpassing OpenAI’s unreleased GPT-4o image generation capabilities. The technology has vast potential for developers, marketers, and businesses, enabling AI-powered design, automated branding, and interactive visual storytelling. Available through Google AI Studio and the Gemini API, Gemini 2.0 Flash is set to redefine AI-assisted creativity and enterprise applications.

💵 Venture Capital updates

Bessemer Venture Partners has raised $350 million for its second India-focused fund, aiming to invest in AI-driven businesses, fintech startups, and direct-to-consumer brands. The U.S.-based venture capital firm, which has backed companies like Swiggy, Ola, and BigBasket, will deploy the new fund starting mid-2025, continuing its strategy of supporting early-stage Indian startups. This follows its first India fund of $220 million raised in 2021, which helped finance companies such as BoldFit, Pepper Content, and Zopper. Despite a slowdown in startup funding across India in 2024, with total investments dropping to $2.7 billion from $4 billion in 2023, Bessemer remains optimistic about the region’s potential. The move aligns with a broader trend of venture capital firms expanding their presence in India, as seen with Accel’s recent $650 million fund and Stellaris Venture Partners closing a $300 million fund.

🫡 Meme of the day

⭐️ Generative AI image of the day