Struggles with programming languages helped research manager Shan Lu find her calling as a bug hunter. She discusses one bug that really haunted her, the thousands she’s identified since, and how she’s turning to LLMs to help make software more reliable.

How do you generate and test materials that don’t exist yet? Researchers Tian Xie and Ziheng Lu share the story behind MatterGen and MatterSim, AI tools poised to transform materials discovery and help drive advances in energy, manufacturing, and sustainability.

Madeleine Daepp, a Microsoft senior researcher specializing in generative AI's political impact, and Robert Osazuwa Ness, who focuses on the reliability of large language models, delve into AI's role in democracy. They discuss the challenges of disinformation, especially during elections, and emphasize the innovations in Taiwan and India’s voting systems. The conversation also highlights the dual-edged nature of AI—its potential benefits and risks—alongside the necessity for media literacy and innovative solutions to combat misinformation.

Just after his NeurIPS 2024 keynote on the co-evolution of systems and AI, Microsoft CVP Lidong Zhou joins the podcast to discuss how rapidly advancing AI impacts the systems supporting it and the opportunities to use AI to enhance systems engineering itself.Learn more:Verus: A Practical Foundation for Systems Verification | Publication, November 2024SuperBench: Improving Cloud AI Infrastructure Reliability with Proactive Validation | Publication, July 2024BitNet: Scaling 1-bit Transformers for Large Language Models | Publication, October 2023

In this special edition of the podcast, Technical Fellow and Microsoft Research AI for Science Director Chris Bishop joins guest host Eliza Strickland in the Microsoft Booth at the 38th annual Conference on Neural Information Processing Systems (NeurIPS) in Vancouver, British Columbia, to talk about deep learning’s potential to improve the speed and scale at which scientific advancements can be made.

Jindong Wang, a researcher, and Steven Euijong Whang, an associate professor at KAIST and co-author of the ERBench paper, dive into the innovative ERBench project designed to evaluate large language models (LLMs). They discuss leveraging relational databases to tackle inaccuracies and enhance response assessments. The duo highlights the importance of integrity constraints in crafting multi-hop questions, as well as the varied performance metrics needed to ensure model trustworthiness, especially in addressing LLM hallucinations.

Next-token prediction trains a language model on all tokens in a sequence. VP Weizhu Chen discusses his team’s 2024 NeurIPS paper on how distinguishing between useful and “noisy” tokens in pretraining can improve token efficiency and model performance.Read the paperGet the code

Can existing algorithms designed for simple reinforcement learning problems be used to solve more complex RL problems? Researcher Dylan Foster discusses the modular approach he and his coauthors explored in their 2024 NeurIPS paper on RL under latent dynamics.Read the paper

Pranjal Chitale discusses the 2024 NeurIPS work CVQA. Spanning 31 languages and the cultures of 30 countries, this VQA benchmark was created with native speakers and cultural experts to evaluate model performance across diverse linguistic and cultural contexts.Read the paperGet the dataset

Nicole Immorlica, Senior Principal Research Manager at Microsoft Research New England, dives into the fascinating intersection of math, economics, and technology. She shares her journey from physics to theoretical economics, revealing how algorithms and AI are reshaping markets. Immorlica discusses the stable marriage problem and its real-world implications, particularly in medical residencies. She also explores the transformative effects of generative AI on creativity and auction design, emphasizing the balance needed for regulation in this evolving landscape.