⚡️ The Future is Spinning Pens

PLUS: 3D-GPT brings 3D worlds to life from text

Good morning. NVIDIA strikes again, creating a riveting AI agent that uses LLMs to transform our expectations of robot capabilities. Let’s dive right in.

Today’s Highlights:

  • 3D-GPT brings 3D worlds to life from just text

  • Apple scatters big bucks for GenAI integration

  • Strategies on the table to curtail Data Center consumption

DEEP DIVE

Nvidia's Eureka AI Can Teach Robots Complex Skills

NVIDIA has achieved a breakthrough, creating an AI agent named Eureka that guides robots in accomplishing complex skills. Developed using GPT-4, Eureka proves capable of teaching robots tasks to the level of human expertise—exemplified by fast-spinning a pen.

"Eureka is a first step toward developing new algorithms that integrate generative and reinforcement learning methods to solve hard tasks."

— Anima Anandkumar, senior director of AI research at NVIDIA

Where does Eureka shine the most?

  • It unexpectedly excels in generating rewards programs for robots that outclass expert human-written ones in over 80% of tasks. This translates to a significant average performance improvement of more than 50% for the bots.

  • Even more impressive, Eureka doesn't rely on task-specific prompting or predefined reward templates. It intelligently incorporates human feedback to accurately align its results to a developer's vision.

Apart from teaching 'pen spinning', Eureka has successfully prepared robots to complete various other complex tasks like opening cabinets and drawers, catching and throwing balls, and manipulating scissors.

Technologies like Eureka open up thrilling possibilities for developers and industries, pioneering the way to a more interactive and advanced AI-robotics ecosystem.

PUNCHLINES

Stacking the Odds: Oracle brings Nvidia AI stack to its marketplace, giving its customers access to top-of-the-line GPUs.

Flash and Dash: Huawei’s OceanStor A310 challenges IBM's (slower) ESS 3500 by promising to feed Nvidia GPUs four times faster.

Wikipedia Watchdog: AI program SIDE aims to improve Wikipedia’s reliability by verifying its citations.

iInvest: Apple is investing $1 billion a year as it gears to integrate GenAI across its offerings.

TLDR

3D-GPT creates 3D worlds from text: Researchers have unveiled 3D-GPT, an AI that employs multiple agents to interpret textual prompts and create accurate 3D assets. This could revolutionize 3D content creation—giving creators in various industries such as gaming, cinema, and VR a more efficient process.

Eleuther AI introduces math-specific LLM, LLEMMA: Researchers from Eleuther AI release LLEMMA, an open-source LLM designed for mathematical problem-solving, outperforming Google’s Minerva and other platforms. Available in 7B and 34B parameter versions, LLEMMA uses tools and theorems without extra fine-tuning, promising significant potential for future AI research.

Chinese AI firm Zhipu raises $340M amidst geopolitical caveats: China’s OpenAI Challenger, Zhipu AI, revealed that it has raised $340M in funding from local investors despite U.S. restrictions on Nvidia chips exports and investment bans. Backed by tech giants like Alibaba, Tencent, and Xiaomi, Zhipu’s open-source models include conversation AI ChatGLM-6B and foundation GLM-130B.

Strategies to Cut Down Data Center Consumption: With U.S. data center demand set to grow by 10% per year until 2030, 75% of organizations aim to establish infrastructure sustainability programs by 2027. Top strategies to reduce consumption include moving to shared cloud infrastructure, adopting liquid cooling, optimizing hardware for specific workloads, embedding sustainability as a long-term goal, and implementing serverless logging to reduce CPU cycles.

TRENDING TOOLS

📊 DataGPT: Get analyst-grade answers to any question in seconds

🧠 Elephas: Create virtual brains from your documents to save time across various platforms

💬 Impaction AI: Analyze conversational data derived from LLM apps

🤖 Autolab: Construct mundane AI agents for real-world tasks

📸 3D to photo by Dabble Studio: Load a 3D model into your browser and virtually photograph it in any scene

That’s all for today—if you have any questions or something interesting to share, please reply to this email. We’d love to hear from you!

P.S. If you want to sign up for the Supercharged newsletter or share it with a friend, you can find us here.

Reply

or to participate.