AI Breakfast
Posts
OpenAI's Operator/Agent coming in January?

OpenAI's Operator/Agent coming in January?

AI Breakfast
November 15, 2024

Good morning. It’s Friday, November 15th.

Did you know: On this day in 2001, the original Xbox went on sale?

In today’s email:

OpenAI’s Policy Paper Hopes to Persuade US Gov
“Operator” Agent Launch Coming in Jan?
Run open-source AI on a Mac
New Experimental Gemini Model
Robot Learns Surgery By Observation
4 New AI Tools
Latest AI Research Papers

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

^{In partnership with OUTREAD AI}

Read 15-minute simplified summaries of groundbreaking research in tech, economics, psychology and more.

Quick insights, big impact – no heavy reading needed!

_{Thank you for supporting our sponsors!}

Today’s trending AI news stories

Report: OpenAI Drafts Policy Paper With Bold Suggestions For US AI Strategy

Image: SiliconAngle

OpenAI’s new policy paper proposes a targeted U.S. AI strategy, urging the creation of AI-focused economic zones to streamline infrastructure projects, from data centers to renewable and nuclear energy. A proposed National Transmission Highway Act would ease energy and data transmission development, attracting private investment through government-backed energy purchases.

Further, OpenAI advocates a North American AI alliance to consolidate competitiveness against global players like China. Collaboration with public universities is outlined to foster AI research hubs and specialized workforce training. These initiatives, OpenAI contends, promise substantial economic gains, drive job creation, GDP growth, and contributions to semiconductor manufacturing and grid modernization. Read more.

OpenAI is reportedly planning AI agent "Operator" for January launch

OpenAI is preparing to launch "Operator," an AI assistant that can autonomously perform tasks such as coding and travel bookings, set for release in January. Initially available as a research preview and through an API for developers, Operator will concentrate on automating browser-based functions.

CEO Sam Altman sees AI agents as the next stage in AI evolution, optimizing the use of existing models. This development mirrors industry efforts, with companies like Anthropic, Microsoft, and Google advancing similar assistants. These agents facilitate the automation of complex workflows by coordinating subtasks across multiple systems.

OpenAI’s "Project Swarm," an open-source framework, illustrates this vision, enabling assistants to transfer control and execute tasks sequentially, showcasing the growing momentum of multi-agent systems in AI-driven automation. Read more.

You can now run the most powerful open source AI models locally on a Mac

Exo Labs has enabled running powerful open-source AI models like Qwen 2.5 Coder-32B and Nemotron-70B on local Mac M4 devices, bypassing costly cloud infrastructure. By connecting multiple Mac Minis and a Macbook Pro M4, Exo Labs provides a cost-effective, secure, and private alternative for AI model execution.

The open-source Exo software distributes workloads across devices, making high-performance AI accessible to individuals and enterprises. Leveraging the M4’s advanced capabilities, Exo offers a decentralized solution with faster, more efficient AI operations. Exo Labs is expanding with enterprise offerings and a benchmarking site to guide users in optimizing their hardware setups. Read more.

New Experimental Gemini Model

The new experimental Gemini model, "gemini-exp-1114," is now available in Google AI Studio and is making waves in the AI community. Recently ranking #1 in overall performance and vision on the Chatbot Arena, the model has seen significant improvements across several domains, including math, hard prompts, creative writing, and coding.

This surge follows a 40-point leap in its score, now surpassing previous iterations. The model's standout features include top rankings in math, hard prompts, and creative writing, while maintaining a strong presence in coding.

As "gemini-exp-1114" continues to gain traction, it is expected to become accessible via API soon, offering users an opportunity to test its capabilities firsthand and provide feedback on its performance. Read more.

AI-powered robot learned to perform surgery by watching doctors work

Researchers at Johns Hopkins and Stanford have pioneered a method that allows robots to learn surgery by watching video footage of expert surgeons, bypassing the need for tedious hand-coded instructions. Using imitation learning, the robot replicates tasks like suturing and tissue manipulation with precision on par with human surgeons.

The system, built on machine learning models akin to those powering language models like ChatGPT, decodes surgical movements into kinematic data. By training on a vast archive of recorded procedures, the robot adapts autonomously, even correcting its actions if necessary. This breakthrough accelerates the training process, offering a path toward faster, more accurate robotic surgeries and inching closer to fully autonomous operations. Read more.

4 new AI-powered tools from around the web

Tilde | Interpreter models to optimize AI

Interpretability is the most important problem in the world. We build moonshot, applied interpretability solutions to make maximally safe and performant AI.

www.tilderesearch.com

Windsurf Editor by Codeium

Tomorrow's editor, today. Windsurf Editor is the first AI agent-powered IDE that keeps developers in the flow. Available today on Mac, Windows, and Linux.

codeium.com/windsurf

ZEPIC | AI-infused customer engagement & marketing automation platform

Self-serve AI marketing platform that unifies data and helps you run cross-channel personalized campaigns. No tech skills required. Try free today.

www.zepic.com

BuildShip | Visual Low-code Backend Builder

Buildship enables businesses to visually create backend APIs, automate complex workflows, and quickly integrate with any tool or AI. Build with ease of no-code and power of code.

buildship.com