Vanishing Gradients

In this discussion, Vincent Warmerdam, a senior data professional at :probabl, challenges conventional data science approaches with innovative insights. He emphasizes the importance of real-world problem exposure and effective visualization. The conversation dives into framing problems accurately and determining if algorithms truly solve them. Vincent advocates for simple models, discusses the role of UI in data science tools, and examines the potential and limitations of LLMs. He highlights the need for community knowledge sharing through blogging and open dialogue.

Explore insights from Eugene Yan, Bryan Bischof, Charles Frye, Hamel Husain, and Shreya Shankar on building end-to-end systems with LLMs, the experimentation mindset for AI products, strategies for building trust in AI, the importance of data examination, and evaluation strategies for professionals. These lessons apply broadly to data science, machine learning, and product development.

Experts from Amazon, Hex, Modal, Parlance Labs, and UC Berkeley share lessons learned from working with Large Language Models. They discuss the importance of evaluation and monitoring in LLM applications, data literacy in AI, the fine-tuning dilemma, real-world insights, and the evolving roles of data scientists and AI engineers.

Alan Nichol, Co-founder and CTO of Rasa, shares insights on using LLMs in chatbots, the evolution of conversational AI, and the challenges of supervised learning. He emphasizes the importance of balancing traditional techniques with new advancements. The podcast also includes a live demo of Rasa's CALM system, showcasing the separation of business logic from language models for reliable conversational flow execution.

Jason Liu, an independent consultant in recommendation systems, discusses building AI apps, playbook for ML, and avoiding pitfalls. They focus on building terrible AI systems to learn how to prevent failures. The podcast explores consulting in various industries, future tooling, and creating robust AI systems.

Sebastian Raschka discusses developing and training large language models (LLMs) from scratch, covering topics like prompt engineering, fine-tuning, and RAG systems. They explore the skills, resources, and hardware needed, the lifecycle of LLMs, live coding to create a spam classifier, and the importance of hands-on experience. They also touch on using PyTorch Lightning and fabric for managing large models, and reveal insights on techniques in natural language processing models and evaluating LLMs for classification problems.

Omoju Miller, a machine learning expert and CEO of Fimio, shares her vision for transparent and reproducible ML workflows. She discusses the necessity of open tools and data in combating the monopolization of tech by closed-source APIs. Topics include the evolution of developer tools, the importance of data provenance, and the potential of a collaborative open compute ecosystem. Omoju also emphasizes user accessibility in machine learning and envisions a future where everyone can build production-ready applications with ease.

Hugo and Johno discuss the evolution of tooling and accessibility in AI over the past decade, highlighting advancements in using big models from Hugging Face and hi-res satellite data. They delve into the Generative AI mindset, democratizing deep learning with fast.ai, and the importance of UX in generative AI applications. The discussion also covers the skill set needed to be an LLM and AI guru, as well as efforts at answer.ai to democratize LLMs and foundation models.

Allen Downey discusses statistical paradoxes and fallacies in using data, including the base rate fallacy and algorithmic fairness. They dive into examples like COVID vaccination data and explore the challenges of interpreting statistical information correctly. The conversation also covers topics such as epidemiological paradoxes, Gaussian distributions, and the importance of understanding biases in data interpretation for media consumption.

Jeremy Howard (Fast.ai), Shreya Shankar (UC Berkeley), and Hamel Husain (Parlance Labs) join Hugo Bowne-Anderson to talk about how LLMs and OpenAI are changing the worlds of data science, machine learning, and machine learning engineering. Jeremy Howard is co-founder of fast.ai, an ex-Chief Scientist at Kaggle, and creator of the ULMFiT approach on which all modern language models are based. Shreya Shankar is at UC Berkeley, ex Google brain, Facebook, and Viaduct. Hamel Husain has his own generative AI and LLM consultancy Parlance Labs and was previously at Outerbounds, Github, and Airbnb. They talk about How LLMs shift the nature of the work we do in DS and ML, How they change the tools we use, The ways in which they could displace the role of traditional ML (e.g. will we stop using xgboost any time soon?), How to navigate all the new tools and techniques, The trade-offs between open and closed models, Reactions to the recent Open Developer Day and the increasing existential crisis for ML. LINKS The panel on YouTube Hugo and Jeremy's upcoming livestream on what the hell happened recently at OpenAI, among many other things Vanishing Gradients on YouTube Vanishing Gradients on twitter