In partnership with

Good morning. It’s Friday, March 14th.

On this day in tech history: 1994: The Linux kernel version 1.0.0 was released. This marked a significant milestone in the development of the open-source operating system, which has since become a cornerstone of modern computing, powering servers, smartphones (via Android), and countless other devices.

In today’s email:

• Gemini AI Upgrade

• OpenAI Tries to ban DeepSeek

• OpenAI supercharges ChatGPT

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Google upgrades Gemini AI: Smarter robots, personalized search, and free Deep Research access

Google DeepMind has introduced Gemini Robotics and Gemini Robotics-ER, AI models that enhance real-time adaptability and spatial reasoning in robotics. Built on Gemini 2.0, these models allow robots to interpret language, navigate changing environments, and refine motor control. Gemini Robotics-ER adds precision grip calculations and autonomous action generation. Partners already integrating the models include Apptronik, Boston Dynamics, and Agility Robotics.

Google also integrated native image generation into Gemini 2.0 Flash, enabling developers to create visuals directly within Google AI Studio and the Gemini API. Unlike traditional systems, Gemini 2.0 Flash maintains character and scene consistency across multiple images while supporting iterative editing through conversational inputs.

Google has also enabled search history integration for Gemini AI, personalizing responses based on user behavior. Meanwhile, Deep Research, an AI-powered analytical tool, is now free a few times a month, providing structured, step-by-step reasoning for complex queries. Read more.

OpenAI and Google fight AI restrictions while DeepSeek defies commercialization

OpenAI and Google are pushing for sweeping AI policy changes in their submissions to the Trump administration’s AI Action Plan, arguing that restrictive regulations threaten U.S. innovation and economic competitiveness.

OpenAI is calling for a ban on AI models from DeepSeek and other Chinese firms, alleging they operate under Beijing’s oversight and are bound by Chinese data-sharing laws. While it previously accused DeepSeek of illicitly replicating its models, no direct link to Chinese state control has been established, despite Beijing’s growing interest. The company also wants exemptions from state AI laws and copyright restrictions, citing China’s rapid AI development as justification.

Google’s proposal similarly advocates for looser copyright and export controls, warning that excessive regulation could weaken U.S. firms while benefiting foreign competitors.

Meanwhile, DeepSeek is taking a different route. According to the Financial Times, DeepSeek’s billionaire founder has chosen to maintain a research-driven approach rather than aggressively scaling commercial operations. Last month marked a milestone for the company, as its revenues covered ongoing operational costs for the first time. Despite achieving financial sustainability, the Chinese lab remains focused on fundamental development rather than immediate revenue growth. Read more.

OpenAI supercharges ChatGPT with Python analytics and precision prompt testing

OpenAI just gave ChatGPT a serious upgrade, merging AI experimentation with real-time analytics. The Chat Playground is now the Prompts Playground, a more surgical interface for testing, refining, and optimizing prompts. Developers can now pull in live data with integrated web and file search, while sharable configurations turn prompt engineering into a structured, repeatable process.

On the analytics front, ChatGPT’s o1 and o3-mini models now support Python-powered data analysis. That means running regressions, visualizing complex business metrics, and simulating scenarios—all within a chat interface. No need to switch tools—ChatGPT can now crunch numbers alongside generating responses. A starter prompt featuring GPT-4o-mini with the Responses API and web search highlights these new capabilities in action. Read more.

• Nous Research drops an API for AI models that OpenAI and Anthropic won’t build

• Anthropic’s CEO warns: $100M AI secrets can vanish in just a few lines of code

• Peter Thiel-backed Palantir eyes finance—new AI venture targets banks and asset managers

• Moonvalley introduces a ‘clean’ generative video AI model for cinema and advertising

• Chinese pilots defeated by AI in high-intensity, unpredictable air combat simulations

• Freepik launches Image-to-Video feature on Google Veo 2 platform

• Meta mocked for raising “Bob Dylan defense” of torrenting in AI copyright fight

• Snap debuts AI-powered Video Lenses, pushing generative tech into AR

• Alibaba’s R1-Omni refines AI emotion recognition with verifiable rewards

• An overview of the GPU market price and its development

• Anthropic updates API to boost throughput and reduce token usage with Claude 3.7 Sonnet

• Google's NotebookLM gets major update: Gemini 2.0 Integration and enhanced citation features

• Bolt launches URL tool to convert Figma designs into full-stack applications

• Convergence launches DeepWork, its most advanced AI agent to date

• Tech leaders at Anthropic, IBM, and Meta warn that AI is coming for software developer jobs

• SoftBank buys $676M old Sharp plant for its OpenAI collab in Japan

• Sesame, the startup behind the viral virtual assistant Maya, releases its base AI model

• Amazon, Google and Meta support tripling nuclear power by 2050

• Google's new AI assistant PixelSense to launch this year, report says

• "No thanks" Fans respond to Microsoft's new Copilot AI 'gaming coach'

📄 LMM-R1: Empowering 3B LMMs with Strong Reasoning Abilities Through Two-Stage Rule-Based RL

📄 TPDiff: Temporal Pyramid Video Diffusion Model  ​​

📄 Reangle-A-Video: 4D Video Generation as Video-to-Video Translation

📄 CoSTAast: Cost-Sensitive Toolpath Agent for Multi-turn Image Editing

📄 World Modeling Makes a Better Planner: Dual Preference Optimization for Embodied Task Planning

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!