The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)

Aakanksha Chowdhery, a machine learning researcher from Reflection, dives into the future of agentic AI. She critiques the current reliance on post-training techniques, advocating for a transformative approach to pre-training. Aakanksha highlights the need for evolving attention mechanisms and tailored training data to enhance long-term reasoning and planning. She discusses the importance of 'trajectory' training data and the risks of synthetic data, all while underscoring the significance of rigorous evaluation in building agentic models.

Munawar Hayat, a researcher at Qualcomm AI Research specializing in multimodal generative AI, dives into the intricacies of Vision-Language Models (VLMs). He discusses the puzzling issue of object hallucination, revealing why these models often overlook visual elements in favor of language. Munawar also introduces attention-guided alignment techniques and a novel approach to generalized contrastive learning for efficient multi-modal retrieval. He shares insights on the Multi-Human Testbench designed to tackle identity leakage challenges in generative models, bringing clarity to this evolving field.

Zain Asgar, co-founder and CEO of Gimlet Labs, is an expert in efficient AI compute orchestration and heterogeneous inference. He discusses the challenges of handling token-heavy agentic workloads and the need for diverse hardware solutions. Zain elaborates on Gimlet's innovative three-layer architecture for workload disaggregation and LLM-driven optimization. He shares insights on the complexities of networking, the trade-offs in precision, and the future of resource scheduling, all while emphasizing the importance of cost-effective AI operations.

Join Devi Parikh, co-founder of Yutori and a pioneering AI researcher, as she discusses the futuristic concept of proactive web agents that can streamline our online interactions. Devi explains how these agents operate using visual models that leverage screenshots for improved reliability over traditional methods. She reveals insights into Yutori's innovative training process, the challenges of web navigation, and the pivotal role of feedback in enhancing agent performance. Tune in to explore how these advancements could reshape our digital experiences!

Robin Braun, VP of AI at HPE, and Luke Norris, CEO of Kamiwaza, delve into AI's role in transforming smart cities and enterprises. They explore the 'Agentic Smart City' project in Vail, focusing on automating website accessibility and digitizing legacy data. The discussion highlights the significance of private cloud infrastructure, fresh data, and practical use cases that deliver real ROI. Learn how innovative technologies can enhance fire detection and urban monitoring, reducing costs and improving compliance in complex workflows.

Carina Hong, founder and CEO of Axiom, discusses her groundbreaking work on creating an AI mathematician. She highlights the convergence of advanced reasoning capabilities, formal proof languages like Lean, and breakthroughs in code generation as key trends revitalizing AI in mathematics. Carina delves into the challenges of translating informal proofs into machine-verifiable formats, the concept of self-improvement through conjecturing and proving, and the potential applications in high-stakes software verification.

Hung Bui, Technology VP at Qualcomm and former head of VinAI Research, dives into high-efficiency generative AI models. He discusses the transition from complex GANs to streamlined diffusion models, revealing his team's breakthroughs with SwiftBrush and SwiftEdit, which allow rapid text-to-image generation and editing. Hung explains their innovative distillation framework that creates a single-step model from a multi-step teacher, and how this paves the way for real-time, on-device AI. The conversation also touches on privacy in personalized agents and efficiency challenges in AI deployment.

In this conversation, Alexandre Pesant, AI lead at Lovable, delves into vibe coding, a transformative practice in software development. He explains how this approach shifts focus from typing code to expressing intent through higher-level abstractions. Alex shares insights on coding agents and the critical role of context engineering. He discusses Lovable’s evolution from complex agent systems to simpler workflows and back as technologies improved. Listeners also gain perspectives on AI's impact on engineering roles and the importance of robust evaluations in creating successful AI-native development tools.

Kunle Olukotun, a Stanford professor and chief technologist at SambaNova Systems, dives into reconfigurable dataflow architectures for AI. He explains how this innovative approach enhances AI inference by dynamically matching hardware to model dataflows, leading to reduced bandwidth bottlenecks. Kunle also highlights the benefits of fast model switching and efficient multi-model serving, crucial for low-latency applications. Plus, he explores future possibilities of using AI to create compilers for evolving hardware setups, offering insights into significant performance improvements.

Jacob Buckman, co-founder and CEO of Manifest AI, dives deep into the world of long-context transformers. He discusses innovative techniques like windowed attention and the revolutionary Power Retention approach, which melds attention and recurrence for astonishing training speeds. Buckman also shares insights on Manifest AI's open-source tools, Vidrial and PowerCoder, and explores the significance of metrics in measuring context utility. Learn about the balance between state and weight for optimal compute architectures and the future potential of context length in AI.