Happy New Year! It’s January 1st.

Did you know: Fear of the Y2K bug lead to apocalyptic preparations?

In today’s email:

• Run:ai Going Open Source

• IBM’s Quantum Computer

• SmolAgents

• 3 New AI Tools

• Latest AI Research Papers

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

Nvidia to open-source Run:ai, the software it acquired for $700M to help companies manage GPUs for AI

Nvidia has sealed a $700 million deal to acquire Run:ai, a platform designed to streamline GPU orchestration for AI workloads. As part of its strategy, Nvidia will open-source Run:ai's technology, expanding its reach beyond Nvidia GPUs to the broader AI landscape. The move signals Nvidia’s continued dominance in the AI space, albeit with a careful nod to regulatory concerns—open-sourcing the platform likely helps sidestep antitrust scrutiny while further cementing Nvidia's foothold.

Run:ai, which has been a trusted Nvidia partner since 2020, promises increased flexibility and efficiency for AI teams managing complex infrastructure, both on-prem and in the cloud. Read more.

IBM will release the largest ever quantum computer in 2025y

IBM is poised to redefine quantum computing with its ambition to develop the largest quantum computer by 2025. The company plans to achieve this by integrating multiple quantum processors into a unified modular system, circumventing the constraints of single-chip architectures. Currently, IBM’s Condor chip boasts 1121 qubits, but the company is preparing to surpass this milestone with its upcoming Flamingo chip.

The modular approach addresses the critical challenge of scaling qubits while maintaining efficient interconnections between chips and their respective wiring. Looking ahead, IBM envisions a network of Kookaburra chips that could reshape both commercial and academic landscapes. The key to this leap lies in the refinement of specialized couplers, essential components that enable seamless communication between qubits and chips over varying distances, ultimately paving the way for fully modular, scalable quantum systems. Read more.

Hugging Face Simplifies AI Agent Development with SmolAgents

Hugging Face has streamlined AI agent creation with SmolAgents, a toolkit that lets developers deploy intelligent agents in just three lines of code. By tapping into Hugging Face's pre-trained models and advanced NLP capabilities, SmolAgents transforms what used to be a daunting, complex process into a simple, modular framework.

It supports everything from rapid prototyping to full-scale production, enabling seamless integration with Hugging Face’s ecosystem for tasks like data retrieval, summarization, and on-the-fly code execution. This toolkit’s lightweight design makes it an ideal choice for developers with limited resources, while its efficiency keeps the focus on solving real-world problems, not wrestling with technical overhead. Read more.

• Meta AI Introduces Mender for Preference-Discerning Models

• Researchers from MIT, OpenAI, Sakana Propose Automating Artificial Life Discovery with ASAL Algorithm

• ByteDance Research Introduces 1.58-bit FLUX: A New AI Approach that Gets 99.5% of the Transformer Parameters Quantized to 1.58 bits

• This AI Paper from Tencent AI Lab and Shanghai Jiao Tong University Explores Overthinking in o1-Like Models for Smarter Computation

• LG debuts new Gram laptop series with onboard small language model

• AgiBot unveils world’s largest humanoid robot training dataset

• Samsung invests $180M+ in Rainbow Robotics to become largest shareholder

• OpenAI co-founder says new AI safety approach "may apply to AGI and beyond"

• ByteDance plans to invest billions in Nvidia AI chips through offshore data centers

• Predicting patients’ sentiments about medications using artificial intelligence techniques

• Researchers successfully train robots to perform surgery by watching videos

• Godfather of AI Throws Support Behind Elon Musk's Lawsuit Against OpenAI

• Tom Cruise's 'Edge of Tomorrow'-like exoskeleton tested in real-world situation

📄 The Open Source Advantage in Large Language Models (LLMs)

📄 Towards Understanding Systems Trade-offs in Retrieval-Augmented Generation Model Inference ​​

📄 The Key of Understanding Vision Tasks: Explanatory Instructions

📄 Bringing Objects to Life: 4D generation from 3D objects

📄 Efficiently Serving LLM Reasoning Programs with Certaindex

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!