(T) Following were my two favorite sessions from Nvidia GTC that were not about Nvidia’s technology or the biotech industry. First, is a presentation from Arthur Mensch, who explains that smaller LLMs, with smarter training, and smarter architectures are the winners. Second, is a fun presentation from Xavier Amatriain, that describes how recommender systems have evolved in particularly recently as LLMs are by themselves recommender systems.

• Mistral AI: Frontier AI in Your Hands [S62450]
  • Arthur Mensch, Chief Executive Officer, Mistral AI
  • “Mistral AI trains state-of-the-art generative models with a strong emphasis on customization and control. It has released the best open-source models as of today. In this keynote, we’ll reflect on the scientific lessons we learned while training our first models (Mistral and Mixtral) and give a glimpse of what’s ahead for 2024.”
  • Key takeways:
    • There is a limit to the number of parameters for LLM
    • Smaller models, smarter architecture, and smarter training are the keys to small and more powerful LLMs

• From Netflix Recommendations to Conversational Multi-agents: The (R)Evolution of AI-Driven Product Innovation [S61329]
  • Xavier Amatriain, VP of Product, Core ML/AI, Google
  • “In an era where ‘Artificial Intelligence’ is not just a buzzword but a core component of every product roadmap, how do we navigate this new central role of AI in product development? As an “AI insider”, Xavier Amatriain has witnessed firsthand the seismic shifts brought about by the AI revolution, especially with the advent of Generative AI and LLMs. This talk will explore AI’s evolution from a mere feature to a ‘product-defining’ element, illustrated by case studies ranging from Xavier’s work leading personalized recommendations at Netflix to the central role of AI as a member of the healthcare team at Curai Health. Moreover, Xavier will guide you through the crucial recalibration needed in the relationship between product teams and AI in this new era where AI is not only the engine behind the product, but becomes the product itself. Join us in exploring how this shift is not just redefining products, but also the very nature of innovation and collaboration in the age of AI.”
  • Key takeways:
    • Recommender systems have evolved from features ETLs for users and items, to feature embeddings
    • LLMs are changing the way to build recommender systems. No need for cosine similarity, just ask the LLM.

Note 1: I will update that blog post with the link to the video when Nvidia will make it available on YouTube.

Note 2: The picture above are California native plants.

Copyright © 2005-2024 by Serge-Paul Carrasco. All rights reserved.
Contact Us: asvinsider at gmail dot com.