Join PhD student Eric Nguyen from Stanford University as he delves into long context language models, discussing Hyena for handling long sequences, FFT optimizations, and the evolution into Hyena DNA and Evo models. Explore the applications in DNA analysis, genetic engineering, and bioinformatics, showcasing the potential of language models in designing DNA for cancer treatment and gene editing with CRISPR-Cas technology.

Andres Ravinet, Sustainability global black belt at Microsoft, discusses AI-driven solutions for sustainability, including early warning systems for extreme weather, reducing food waste, conserving the Amazon rainforest, and standardized sustainability compliance reporting. He also explores the application of LLMs and generative AI in addressing sustainability challenges.

Fatih Porikli from Qualcomm AI Research discusses their research at CVPR, covering topics like text-to-image generation, video reasoning using language models, real-time 360° image generation, visual reasoning for mathematical plots, and more. They also touch on demos like multi-modal vision-language models and parameter-efficient fine tuning for mobile phones.

Sasha Luccioni, AI and Climate lead at Hugging Face, discusses the energy consumption of AI models. She compares the efficiency of pre-trained models vs. task-specific models, highlighting the implications and challenges. Sasha introduces Energy Star Ratings for AI Models as a system to select energy-efficient models. The discussion explores the environmental impact, challenges in evaluation, and the importance of documentation standards for AI models.

In this episode, Christopher Manning, a leading expert in NLP and Machine Learning, discusses the intersection of linguistics and large language models, the intelligence of LLMs, and the future of AI research. He explores the reasoning capabilities of LLMs, shares insights on alternative architectures beyond LLMs, and highlights opportunities in AI research.

Machine learning scientist Abdul Fatir Ansari discusses 'Chronos: Learning the Language of Time Series,' covering challenges in leveraging pre-trained language models, advantages over statistical models, zero-shot forecasting benchmarks, improving synthetic data quality, and potential production system integration.

Joel Hestness, from Cerebras, discusses the Wafer Scale Engine 3 for large language models, software support for ML frameworks, and research on weight-sparse training. The episode explores the unique design of WSE chip, optimizations for AI clusters, and applications in medical, finance, and civil support services.

Laurent Boinot from Microsoft discusses AI's role in power systems, including demand forecasting & grid optimization. AI ensures secure systems, customer interactions, efficient knowledge bases, & electric transmission designs. Also delves into future of nuclear power & importance of electric vehicles in American energy management.

Research scholar Aza Jalalvand discusses using deep reinforcement learning to control plasma instabilities in fusion reactors. Topics include detecting 'tearing mode' instability, collecting and processing complex diagnostic data, training models, deploying controller algorithms, and future opportunities for AI in fusion energy production.

Kirk Marple, CEO of Graphlit, discusses GraphRAG architecture, entity extraction, prompt compilation for LLMs, and use cases. The podcast delves into the integration of multi-modal data, challenges in data storage models, and future agent-based applications enabled by the approach.