Latent Space: The AI Engineer Podcast

In this engaging discussion, Jonathan Frankle, Chief AI Scientist at Databricks, and Josh Albrecht, CTO of Imbue, dive into groundbreaking advancements in AI. They unveil Imbue 70B, a model outperforming GPT-4o with significantly less data. The duo shares insights on the complexities of scaling GPU clusters and the importance of high-performance infrastructure. They also address evaluating language models and introduce innovative tools for hyperparameter tuning. Their expertise shines through as they explore the future of AI in coding and reasoning tasks.

The World’s Fair for AI engineers is sold out, generating excitement in the community. Discussions revolve around the evolving role of AI engineers, distinguishing them from traditional ML engineers. Key insights include the necessity for specialization and new skill sets. The podcast also explores collaborative dynamics between engineers and product managers, and the impact of vertical AI startups. Trends in AI inference and creativity highlight innovative potentials, setting the stage for emerging technologies and community engagement.

James Brady, Head of Engineering at Elicit, and Adam Wiggins, Internal Journalist at Elicit and co-founder of Heroku, dive into the world of hiring AI engineers. They explore the blend of traditional and AI-focused skills needed in candidates and share their journey transitioning into AI roles. The duo discusses the complexities of error handling, interview strategies, and balancing innovation with user experience. They emphasize the importance of community connections and employer branding to attract top talent in the fast-evolving AI landscape.

Mike Conover, Founder of Brightwave and former leader of the OSS models team at Databricks, shares insights on the transformative role of AI in finance. He discusses how language models reflect societal beliefs and the significant funding Brightwave secured to enhance market understanding. The conversation also dives into Brightwave's capabilities in combating information overload for finance professionals. They explore challenges in data handling and the balance between AI and human decision-making, particularly in hedge funds. A fascinating look at the future of investment analytics!

Expert guests Graham Neubig and Aman Sanger discuss AI topics like Code Edits, Sandboxes, Academia vs Industry. They delve into Benchmarks like SWEBench, Dataset Contamination Detection, and GAIA Benchmark. The conversation also touches on Reasoning - Self-RAG, Let's Verify Step By Step, and developments in multi-agent systems with MetaGPT.

Mark Huang, Co-founder of Gradient.ai, dives into the exciting advancements in AI, particularly long context learning. He discusses the evolution of context lengths, mapping out a timeline of breakthroughs and innovations in LLMs. Mark reflects on his team's work with Llama 3 and the challenges of training models with vast token capacities. He also sheds light on optimizing GPU performance and the pressing need for high-quality data in model training. Their vision is creating flexible AI solutions that truly adapt to enterprise workflows.

Christian Szegedy, Ilya Sutskever, and Durk Kingma discuss the most notable topics from ICLR 2024, including expansion of deep learning models, latent variable models, generative models, unsupervised learning, adversarial machine learning, attention maps in vision transformers, efficient model training strategies, and optimization in large GPU clusters.

Jesse Silver, co-founder of a platform empowering OnlyFans creators with AI chatbots, explores the lesser-known aspects of adult content technology. He discusses how NSFW chatbots enhance fan interactions and boost revenue while tackling the challenges of maintaining brand voice. The growth of 'AI waifus' has reshaped engagement, with intriguing insights into prompt injections and the importance of safety in AI conversations. Jesse emphasizes the innovative potential in this field, especially in enhancing the connection between creators and their audiences.

In this engaging discussion, Joscha Bach, an AI researcher specializing in cognitive architectures, and Karan Malhotra, CEO of Nous Research, explore the untapped potential of simulative AI. They dive into interactive AI capabilities, highlighting WebSim's creative tools and alternate realities. The duo also examines the evolving relationship between AI and creativity, touching on ethical considerations and the quest for consciousness in AI systems. With insights into the challenges of language models and community dynamics in AI, the conversation offers a captivating glimpse into the future of technology.

Jason Liu, the creator of the Instructor library and former machine learning engineer at Stitch Fix, discusses his journey from skepticism to embracing AI technologies. He shares insights on the evolution of prompt engineering, revealing how his tool simplifies interactions with AI and enhances JSON handling. The conversation delves into the challenges of tech entrepreneurship, emphasizing personal agency and lessons from failures. Liu also highlights the impact of structured outputs in AI development and the growing importance of flexibility in tech projects.