Practical AI

Worlds are colliding! This week we join forces with the hosts of the MLOps.Community podcast to discuss all things machine learning operations. We talk about how the recent explosion of foundation models and generative models is influencing the world of MLOps, and we discuss related tooling, workflows, perceptions, etc.Join the discussionChangelog++ members save 2 minutes on this episode because they made the ads disappear. Join today!Sponsors:Fastly – Our bandwidth partner. Fastly powers fast, secure, and scalable digital experiences. Move beyond your content delivery network to their powerful edge cloud platform. Learn more at fastly.comFly.io – The home of Changelog.com — Deploy your apps and databases close to your users. In minutes you can run your Ruby, Go, Node, Deno, Python, or Elixir app (and databases!) all over the world. No ops required. Learn more at fly.io/changelog and check out the speedrun in their docs. Changelog++ – You love our content and you want to take it to the next level by showing your support. We’ll take you closer to the metal with extended episodes, make the ads disappear, and increment your audio quality with higher bitrate mp3s. Let’s do this! Featuring:Demetrios Brinkmann – XMihail Eric – Website, GitHub, XChris Benson – Website, GitHub, LinkedIn, XDaniel Whitenack – Website, GitHub, XShow Notes:MLOps.CommunitySomething missing or broken? PRs welcome!

Beau Perschall, Director of Omniverse Sim Data Ops at NVIDIA, dives into the cutting-edge world of 3D environments and their role in simulation for AI. He demystifies the Omniverse, explaining how it combines 3D tech with AI for real-world applications like robotics and autonomous vehicles. The conversation highlights the significance of synthetic data and how tools like SimReady streamline the preparation of 3D assets. Beau also touches on the future of AI in this space, emphasizing collaboration across sectors to unlock its full potential.

Nader Khalil, Co-founder and CEO of brev.dev, dives into the hassle of creating reproducible GPU development environments for AI. He shares insights on overcoming challenges in development settings, especially during the pandemic. The conversation covers advantages of cloud-based solutions for cost efficiency and improved user experience. Khalil also discusses the future of AI tools, emphasizing seamless integration and customizable environments to enhance productivity and innovation in GPU deployment.

Kirsten Lum, Co-founder and CPO of Storytellers AI, shares her insights on the unique challenges small organizations face in implementing machine learning. She highlights that underutilization often stems from mismanagement, not a lack of data. Kirsten discusses the importance of adaptable data scientists and effective project management in driving successful ML practices. Additionally, she emphasizes the shift towards valuing real-world outcomes over mere model accuracy, advocating for a cultural change in data-driven decision-making.

A fascinating exploration of OpenAI’s ChatGPT reveals how this groundbreaking language model is reshaping our world. The hosts discuss its training process using human feedback and its impact on creative industries and storytelling. Personal anecdotes highlight its role in everyday life, emphasizing the synergy between human creativity and AI. As advancements continue, the future promises even more intriguing developments, making AI a crucial player in various sectors from entertainment to nonprofits.

Just Zwennicker, a data engineer who created a machine translation system for Sranan Tongo, and Bonaventure Dossou, a Ph.D. student focusing on low-resource languages, explore the importance of NLP for local language communities. They discuss the challenges of data scarcity and the need for representation in technology, particularly for African languages. The conversation highlights grassroots initiatives like Masakane, which empowers native speakers to enhance linguistic diversity in NLP systems, ensuring their languages thrive in the digital age.

José Souza and Ricardo Rei from Unbabel dive into the cutting-edge world of machine translation. They discuss the importance of quality estimation and innovations like COMET, which blends human expertise with automated evaluation. The duo highlights advancements in translation assessment methods, moving beyond traditional metrics like BLEU to more accurate systems. They also explore the future of translation technology and the challenges of integrating quality predictions, making it a fascinating listen for anyone interested in AI's impact on language.

Matthew Lanham, Academic Director at Purdue University, and Mark Tabladillo, a Cloud Architect at Microsoft, dive into the exciting world of AI competitions. They discuss how over 170 student teams tackled image captioning in multiple languages, showcasing their innovative strategies. The duo emphasizes the evolving role of cloud resources in data science and the educational impact of these competitions. They also explore the challenges of diverse linguistic datasets and how AI can help preserve cultural heritage while promoting business innovation.

The podcast tackles the lawsuit against GitHub, Microsoft, and OpenAI over Copilot's code suggestions, questioning if it breaches open source licenses. It also dives into concerns about Meta AI's Galactica model, which has produced alarming science-related outputs. The hosts discuss the ethical complexities of AI tools, emphasizing the need for regulation and understanding of copyright issues. They highlight the necessity of accuracy and trust in AI, underscoring how these technologies must evolve responsibly to prevent misinformation.

Matar Haller, VP of Data at ActiveFence, leads efforts to combat online harms like disinformation and extremism using AI and expert insights. He shares the importance of adapting technologies for ongoing challenges, especially in online safety for children and effective content moderation. The conversation highlights how robust safety measures can enhance user trust and discusses the intricate balance of using data-driven models to identify harmful content. Matar also touches on the emotional challenges of dealing with distressing material in their mission-driven work.