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

From Math Olympiads to Navier-Stokes: How Fast Is AI Progressing? with Greg Burnham - #778

From Voice Agents to AI Avatars with Alexander Smola - #777

Voice AI is evolving from brittle, text-based interfaces toward fluid, multimodal systems capable of real-time interaction and emotional intelligence. Alex Smola, CEO of Boson AI, emphasizes that achieving natural dialogue requires balancing low-latency inference with high-fidelity audio processing. Rather than relying...

Do AI Tokenomics Matter More Than Model Benchmarks? with Chris Potts - #776

The economics of AI, or "tokenomics," requires a shift from measuring model benchmarks to evaluating the return on investment for token consumption. As AI providers test pricing models, users face significant cost increases, necessitating a "consumer price index" for engineering tasks to track value beyond simple pull ...

World Models and the Future of Spatial AI with Justin Johnson - #775

World models represent the next frontier in AI, shifting from simple language generation to systems capable of simulating, reconstructing, and acting within complex environments. Justin Johnson, co-founder of World Labs and Associate Professor at the University of Michigan, clarifies that while the field lacks a singul...

Why the Next AI Breakthrough May Come from Physics with Max Welling - #774

Max Welling, a researcher and entrepreneur, applies geometric deep learning and equivariance to materials science through his startup, CuspAI. By utilizing generative models and molecular dynamics, the company accelerates the discovery of materials for carbon capture, semiconductors, and energy storage, effectively rep...

Why Image Generation Needs More Than Bigger Models with Fatih Porikli - #773

Text-to-image generation has evolved from a frontier research problem into a highly capable technology, yet significant challenges remain regarding controllability, identity consistency, and computational efficiency. Fatih Porikli, Vice President of Technology at Qualcomm, highlights that current models often struggle ...

Why Models Are AI’s Next Training Dataset with Damian Borth - #772

Weight space learning treats the parameters of trained neural networks as a primary data modality, enabling researchers to analyze, compress, and generate new models by learning from existing weight configurations. Rather than viewing weights solely as the final product of training, this approach leverages them as inpu...

How AI Learns to Smell with Alex Wiltschko - #771

Digitizing the sense of smell requires translating molecular structures into digital information through olfactory intelligence. Alex Wiltschko, founder and CEO of Osmo, leverages graph neural networks to model, predict, and design scents, effectively bridging the gap between chemistry and computing. By training models...

Why AI Agents Break the GenAI Security Model with Devvret Rishi - #770

Autonomous AI agents introduce significant security risks in enterprise environments because they operate faster than human oversight and can creatively circumvent static guardrails. Conventional security models, which rely on manual approval and deterministic rules, fail to manage the speed and complexity of these age...

Is RAG Dead? Lessons from Building AI for Tax Law with Alex Bowcut - #769

Retrieval-Augmented Generation remains essential for high-stakes domains like tax compliance where accuracy and verifiable citations are non-negotiable. Sphere’s Tax Review and Assessment Model (TRAM) demonstrates this by automating the analysis of complex, global legislative documents. The system employs a sophisticat...

Relational Foundation Models for Enterprise Data with Jure Leskovec - #768

How to Find the Agent Failures Your Evals Miss with Scott Clark - #767

How to Engineer AI Inference Systems with Philip Kiely - #766

Inference engineering represents the most critical and complex workload in the AI stack, requiring a multidisciplinary approach that blends GPU-level programming, distributed systems architecture, and the rapid application of emerging research. As AI models scale, the timeline for moving research into production has co...

How Capital One Delivers Multi-Agent Systems with Rashmi Shetty - #765

Multi-agentic AI systems represent a shift from simple LLM-based text generation to goal-oriented, autonomous action-taking within complex enterprise environments. Rashmi Shetty, Senior Director of Enterprise Generative AI Platform at Capital One, details how the organization leverages multi-agent architectures to solv...

The Race to Production-Grade Diffusion LLMs with Stefano Ermon - #764

Agent Swarms and Knowledge Graphs for Autonomous Software Development with Siddhant Pardeshi - #763

AI Trends 2026: OpenClaw Agents, Reasoning LLMs, and More with Sebastian Raschka - #762

The podcast explores the current state and future trends of Large Language Models (LLMs), particularly focusing on advancements since last year and expectations for 2026. Independent LLM researcher Sebastian Raschka highlights the shift towards post-training techniques, such as reasoning and tool use, as key areas of d...

The Evolution of Reasoning in Small Language Models with Yejin Choi - #761

Intelligent Robots in 2026: Are We There Yet? with Nikita Rudin - #760

Rethinking Pre-Training for Agentic AI with Aakanksha Chowdhery - #759