In partnership with

Good morning. It’s Friday, January 17th.

Did you know: On this day in 1984, the U.S. Supreme Court ruled in a 5-4 decision that the personal taping of television programs using a home VCR to watch later is considered fair use and is not copyright infringement. It was known as the "Betamax case."

In today’s email:

• China’s Ultrafast Robot Dog

• Luma AI's Ray2

• Microsoft's MatterGen

• 3 New AI Tools

• Latest AI Research Papers

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

China's robot dog breaks world record by running 100 meters under 10 seconds

China’s Black Panther 2.0 robot dog has shattered speed records, completing 100 meters in under 10 seconds. The result of a collaboration between Zhejiang University and startup Mirror Me, this quadruped is inspired by the biomechanics of black panthers and jerboas. Its carbon-fiber shins enhance rigidity by 135%, while cheetah-like shoes amplify grip by 200%.

Equipped with shock-absorbing springs at key joints and driven by AI, the robot adapts its gait to varying terrains, optimizing speed and fluidity. Weighing 83 pounds, the Black Panther 2.0 now surpasses previous robotic speed records, though it remains slower than top animal sprinters. The team aims to push its agility further, potentially positioning the robot for industrial and practical applications. Read more.

Luma AI's Ray2 Surpasses Expectations in Motion and Physics

Luma AI’s Ray2 has arrived, retooling the video generation landscape with speed and lifelike motion. With ten times more compute power than its predecessor, Ray1, it pushes the limits of what’s possible, producing video that feels fluid and realistic—without the slow-motion pitfalls of competitors. For now, the model’s text-to-video capability lets users generate 5-10 second clips in mere seconds, though the high demand might cause brief delays.

Early tests show an impressive range, from dramatic snowstorms to surreal surfing giraffes, all with natural movement. Luma’s roadmap includes adding image-to-video and editing features soon, further enhancing creative potential.

To celebrate, Luma is hosting the Ray2 Awards, offering creators the chance to win cash prizes and earn commissions through an affiliate program. Read more.

Microsoft's MatterGen is a new AI Reinventing Material Discovery

Microsoft Research introduces MatterGen, an AI model that shifts the paradigm in material design. Unlike traditional methods that painstakingly sift through existing compounds, MatterGen generates entirely new materials by refining random atomic arrangements, ensuring they meet desired criteria.

Using a diffusion model tailored for three-dimensional crystal structures, the system has outperformed predecessors in both stability and novelty. Its potential spans industries from energy storage to semiconductor development, with applications that could accelerate advancements in electric vehicles and renewable energy.

Microsoft’s open-source release of the model allows global collaboration, promising rapid innovation. However, the road from simulation to real-world application remains long, with testing and validation required for industrial-scale deployment. Read more.

• OpenAI is bankrolling Axios' expansion into four new markets

• Google is making AI in Gmail and Docs free — but raising the price of Workspace

• Microsoft Launches Pay-Per-Message AI Pricing and Free Copilot Chat Tier

• François Chollet bets on AGI with Ndea’s program synthesis vision

• Zuckerberg compares Meta's AI data use to YouTube's piracy fight

• Microsoft’s AutoGen Update Powers Scalable, Distributed AI Agent Networks

• LinkedIn's New and Free AI Tools Tackle Recruitment Bottlenecks

• Hinge’s new AI feature determines if your prompt response is too basic

• GPT-4 After-School Tutoring in Nigeria Delivers 2 Years of Learning Gains in Just 6 Weeks

• Frequent AI use linked to lower critical thinking scores, research shows TikTok owner

• ByteDance powered an e-reader's unhinged AI assistant

• Singapore Chip-Gear Maker GVT Says AI Boom Far From Over

• Nvidia sharpens AI guardrails with new safety microservices

• Physical Intelligence's (π) New Tokenizer Enables 5x Faster Training for Embodied AI

• Sakana AI’s Transformer² is an AI Model That Adapts and Learns Like a Living Brain

• Black Forest Labs API Enables Flux Pro Model Finetuning with Minimal Samples

• AI Bot Secures $8M Seed Round, Bypassing Human Involvement Until Contract Stage

• Mistral signs deal with AFP to offer up-to-date answers in Le Chat

• Google’s New LLM Architecture Splits Memory to Lower Capacity and Compute Costs

• Apple is pausing notification summaries for news in the latest iOS 18.3 beta

• Unitree G1 Bionic: Agile Upgrade

• New AI tool detects fake news with 99% accuracy

• KREA AI launches free 3D object conversion tool for image transformation

📄 MMDocIR: Benchmarking Multi-Modal Retrieval for Long Documents

📄 Imagine while Reasoning in Space: Multimodal Visualization-of-Thought ​​

📄 OmniThink: Expanding Knowledge Boundaries in Machine Writing through Thinking

📄 Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps

📄 Towards Best Practices for Open Datasets for LLM Training

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!