Chuck Byers with OMG

Chuck Byers, a seasoned industry professional, discussed the integration of AI and IoT technologies in the context of the Object Management Group (OMG) and the Digital Twin Consortium. Byers highlighted the importance of AI in real-time systems, emphasizing the need for prompt engineering and meta AI to ensure trustworthiness. He detailed the computational requirements of training large language models, noting that retraining the ChatGPT 3 model would require 195 NVIDIA DGX servers, costing $80 million and consuming 2.5 million watts. Byers also stressed the importance of sustainable energy sources and efficient cooling solutions for future data centers.

Action Items

• [ ] Explore ways to detect when AI results are not as trustworthy as they need to be, and develop methods to validate AI outputs
• [ ] Investigate the use of small modular nuclear reactors to power data centers for AI model training
• [ ] Promote the work of OMG and the Digital Twin Consortium in addressing the challenges of AI and the Internet of Things

Outline

Introduction and Welcome

• Scott MacKenzie introduces the podcast and its focus on industry innovations and trends.
• Scott welcomes listeners and mentions the broadcasting location at the OMG SQ1 meeting in Reston, Virginia.
• Scott introduces Chuck Byers, highlighting his extensive background in electrical engineering and his contributions to Bell Labs and Cisco.
• Chuck Byers joins the conversation, expressing his pleasure in being invited and his admiration for Scott's hosting style.

Chuck Byers' Background and Contributions

• Chuck Byers shares his educational background, including a Master's degree in electrical engineering and teaching at the University of Wisconsin.
• He discusses his work at Bell Labs and his invention of a US patent for a device that alerts users to approaching badness.
• Chuck talks about his tenure at Cisco, where he worked on computer control, IoT, edge computing, and drone technology.
• He mentions his current role as Chief Technical Officer of the Industry IoT Consortium and its recent integration into the Digital Twin Consortium.

Digital Twin Consortium and Industry IoT Consortium

• Chuck explains the Digital Twin Consortium's focus on pairing simulation models with real-time physical systems.
• He describes the Industry IoT Consortium's focus on sensors, actuators, and the computation necessary for IoT systems.
• Chuck highlights the synergy between the two consortiums, emphasizing their combined potential to sense, network, and actuate the physical world into the digital world.
• He provides an example of using simulation models to optimize oil refinery processes without physical experimentation.

Challenges and Opportunities in AI

• Scott MacKenzie and Chuck Byers discuss the challenges of AI, particularly the lack of standards and trustworthiness.
• Chuck explains the different types of AI, including machine vision and generative AI, and the unique challenges of generative AI.
• He describes the use of large language models in generative AI and their applications in tools like ChatGPT and Microsoft 365.
• Chuck emphasizes the importance of prompt engineering and fact-checking to ensure the accuracy of AI-generated outputs.

Meta AI and AI Trustworthiness

• Chuck introduces the concept of meta AI, which involves using multiple independent AI systems to cross-validate results.
• He explains how meta AI can help filter out biases and hallucinations by comparing results from different AI models.
• Scott mentions a platform called h2o.ai that queries multiple AI engines and displays results, though it still requires human oversight.
• Chuck discusses the potential of automating this process to achieve real-time, trustworthy AI results.

Computational Power and Energy Efficiency

• Chuck delves into the computational power required to train generative AI models, citing the example of the ChatGPT 3.5 model.
• He explains the role of GPUs in handling the massive computational requirements and the efficiency of NVIDIA's GPGPU technology.
• Chuck calculates the cost and energy consumption of training a generative AI model using NVIDIA DGX servers.
• He highlights the importance of sustainable energy sources and efficient cooling solutions for data centers.

Future of Data Centers and AI

• Chuck outlines the future of data centers, including the use of renewable energy, small modular reactors, and immersion cooling.
• He discusses the compact, fiber-optic interconnected data centers of the future and their potential cost.
• Chuck emphasizes the need for efficient energy use and cooling solutions to support the growing demand for AI computational power.
• He predicts that these advancements will make data centers more sustainable and cost-effective.

Conclusion and Contact Information

• Scott MacKenzie expresses his hope for the success of AI and the role of organizations like OMG in defining standards.
• Chuck Byers reiterates the importance of trustworthiness, sustainability, and affordability in AI development.
• Scott invites listeners to engage with OMG and contact Chuck for more information on AI and the consortium.
• The conversation concludes with Scott encouraging listeners to educate themselves and get involved in the industry.