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

Dan Friedman, a PhD student from Princeton's NLP group, dives into his fascinating research on mechanistic interpretability for transformer models. He discusses his innovative paper that modifies transformer architecture to create human-readable programs. The conversation uncovers the challenges of current interpretability methods and contrasts them with his approach. They explore the RASP framework's role in transforming programs and delve into the complexities of optimizing model constraints, highlighting the importance of clarity in understanding AI.

Thomas Dietterich, a distinguished professor emeritus at Oregon State University, dives into the latest trends in AI and machine learning. He discusses the strengths and weaknesses of large language models like GPT-4, while exploring their potential limitations in reasoning. The conversation covers topics like uncertainty quantification and the fascinating world of 'hallucinations' in language models. Dietterich also offers predictions for 2024 and motivates newcomers to tap into the field's endless possibilities.

Naila Murray, Director of AI Research at Meta, discusses the cutting-edge landscape of computer vision. They explore advancements like controllable AI generation, multimodal models, and tools such as Segment Anything for intuitive image segmentation. Naila dives into the possibilities of ControlNet and DINOv2, emphasizing their roles in object recognition and complex scenarios. Looking ahead, she shares insights on opportunities in self-supervised learning and generative models, forecasting exciting innovations for 2024 in AI.

Joining the conversation is Ed Anuff, Chief Product Officer at DataStax, who brings his extensive experience in startups and technology. He delves into the fascinating world of vector databases, discussing their critical role in handling massive, unstructured datasets. Ed highlights advancements in algorithms like HNSW and explores how embedding models enhance database retrieval. He shares insights on integrating live data into AI applications, the significance of data chunking, and the potential of GPUs to boost performance in generative AI systems.

In this discussion, Markus Nagel, a research scientist at Qualcomm AI Research, shares insights from his recent papers at NeurIPS 2023, focusing on machine learning efficiency. He tackles the challenges of quantizing transformers, particularly in minimizing outlier issues in attention mechanisms. The conversation explores the pros and cons of pruning versus quantization for model weight compression and dives into innovative methods for multitask and multidomain learning. Additionally, the use of geometric algebra in enhancing algorithms for robotics is highlighted.

Michael Kearns, a professor at the University of Pennsylvania and Amazon scholar, dives into the new challenges of responsible AI in the generative era. He discusses the evolution of service card metrics and their limitations in evaluating AI performance. Kearns also tackles the complexities of evaluating large language models and introduces the concept of clean rooms in machine learning, emphasizing privacy through differential techniques. With insights from his work at AWS, he advocates for collaboration between AI developers and stakeholders to enhance ethical practices.

Mike Miller, Director of Product at AWS, leads the charge in developing engaging AI edutainment tools. He dives into AWS PartyRock, a playful, no-code generative AI app builder, making app creation fun and accessible. The conversation highlights innovations like DeepRacer, an RC car navigating AI challenges, and DeepLens, a groundbreaking computer vision tool. Miller emphasizes the importance of blending education with entertainment, inviting listeners to unleash their creativity through intuitive AI-powered applications.

Cody Coleman, co-founder and CEO of Coactive AI, discusses the innovative applications of data-centric AI in building a multimodal asset platform. He delves into active learning and core set selection, explaining how these techniques boost efficiency in machine learning. The conversation also highlights Coactive's use of multimodal embeddings for visual search and the infrastructure optimizations that support scalability. Cody shares insights and advice for entrepreneurs in the generative AI space, making complex topics accessible to all.

Join Kyle Roche, the founder and CEO of Griptape and former GM at AWS, as he dives deep into middleware for generative AI. He unveils innovative patterns for LLM applications, including off-prompt data retrieval and flexible pipeline management. Roche discusses how Griptape enhances data connectivity while addressing privacy and management concerns. Tune in to learn about driving efficiencies in various industries and the impact of responsible AI solutions!

In this discussion, Prem Natarajan, Chief Scientist and Head of Enterprise AI at Capital One, tackles AI access and inclusivity as critical challenges in banking. He highlights the importance of diversity in data sets to combat biases and improve fraud detection. Prem shares insights on the use of foundation models and federated learning, emphasizing data quality and privacy preservation. He also stresses the need for collaboration between academia and industry to enhance AI impact, ultimately advocating for mission-inspired research that benefits customers and the broader community.