GPU Technology Conference Keynote Oct 2020 | Part 4: "AI: Software that Writes Software" | Summary and Q&A

160.7K views
October 5, 2020
by
NVIDIA
YouTube video player
GPU Technology Conference Keynote Oct 2020 | Part 4: "AI: Software that Writes Software"

TL;DR

AI is rapidly advancing, with increasing complexity and capabilities, as shown by the exponential growth in computation required to train state-of-the-art models. NVIDIA is democratizing AI and expanding its reach to more industries and cloud service providers.

Install to Summarize YouTube Videos and Get Transcripts

Questions & Answers

Q: What is the exponential growth in computation required to train state-of-the-art AI models?

The computation needed to train such models has increased 30,000 times in less than five years, doubling every couple of months. This demonstrates the scalability and complexity of AI development.

Q: How does NVIDIA's AI technology perform on reading comprehension tests compared to humans?

NVIDIA's Natural Language Understanding AI achieved a score of 91% on the RACE reading comprehension test, outperforming the average human score of 73%. Although expert humans scored 95%, it demonstrates the high capabilities of AI in this domain.

Q: How has AI been utilized in quantum chemistry?

Researchers at Caltech have successfully used deep learning in quantum chemistry to predict molecular-protein reaction energy. This advancement has accelerated simulations by 1000 times, showing the potential of AI in the field.

Q: What AI-related services does NVIDIA offer for cloud service providers?

NVIDIA provides AI application frameworks, pre-trained models, transfer learning tools, and runtime engines to support self-driving cars, robotics, drug discovery, conversational AI, and recommender systems. Their NGC cloud platform is being integrated into Azure, AWS, and GCP marketplaces.

Summary & Key Takeaways

  • AI is evolving at an unprecedented pace, with models requiring 30,000 times more computation to train in just five years.

  • NVIDIA has developed remarkable AI applications, such as natural language understanding, drone flight control, quantum chemistry prediction, and protein structure prediction.

  • NVIDIA offers three pillars of AI: data processing and training, inference with tools like TensorRT and Triton Inference Server, and AI application frameworks for various industries.

Share This Summary 📚

Summarize YouTube Videos and Get Video Transcripts with 1-Click

Download browser extensions on:

Explore More Summaries from NVIDIA 📚

Summarize YouTube Videos and Get Video Transcripts with 1-Click

Download browser extensions on: