AI Stories – Details, episodes & analysis

Our guest today is Sebastian Raschka, Senior Staff Research Engineer at Lightning AI and bestselling book author.

In our conversation, we first talk about Sebastian's role at Lightning AI and what the platform provides. We also dive into two great open source libraries that they've built to train, finetune, deploy and scale LLMs.: pytorch lightning and litgpt.

In the second part of our conversation,  we dig into Sebastian's new book: "Build and LLM from Scratch". We discuss the key steps needed to train LLMs, the differences between GPT-2 and more recent models like Llama 3.1, multimodal LLMs and the future of the field.

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Build a Large Language Model From Scratch Book: https://www.amazon.com/Build-Large-Language-Model-Scratch/dp/1633437167

Blog post on Multimodal LLMs: https://magazine.sebastianraschka.com/p/understanding-multimodal-llms

Lightning AI (with pytorch lightning and litgpt repos): https://github.com/Lightning-AI

Follow Sebastian on LinkedIn: https://www.linkedin.com/in/sebastianraschka/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

---

(00:00) - Intro

(02:27) - How Sebastian got into Data & AI

(06:44) - Regressions and loss functions

(13:32) - Academia to joining LightningAI

(21:14) - Lightning AI VS other cloud providers

(26:14) - Building PyTorch Lightning & LitGPT

(30:48) - Sebastian’s role as Staff Research Engineer

(34:35) - Build an LLM From Scratch

(45:00) - From GPT2 to Llama 3.1

(48:34) - Long Context VS RAG

(56:15) - Multimodal LLMs

(01:03:27) - Career Advice

Our guest today is Loubna Ben Allal, Machine Learning Engineer at Hugging Face 🤗 .

In our conversation, Loubna first explains how she built two impressive code generation models: StarCoder and StarCoder2. We dig into the importance of data when training large models and what can be done on the data side to improve LLMs performance.

We then dive into synthetic data generation and discuss the pros and cons. Loubna explains how she built Cosmopedia, a dataset fully synthetic generated using Mixtral 8x7B.

Loubna also shares career mistakes, advice and her take on the future of developers and code generation. 

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Cosmopedia Dataset: https://huggingface.co/blog/cosmopedia

StarCoder blog post: https://huggingface.co/blog/starcoder

Follow Loubna on LinkedIn: https://www.linkedin.com/in/loubna-ben-allal-238690152/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

---

(00:00) - Intro

(02:00) - How Loubna Got Into Data & AI

(03:57) - Internship at Hugging Face

(06:21) - Building A Code Generation Model: StarCoder

(12:14) - Data Filtering Techniques for LLMs

(18:44) - Training StarCoder

(21:35) - Will GenAI Replace Developers? 

(25:44) - Synthetic Data Generation & Building Cosmopedia

(35:44) - Evaluating a 1B Params Model Trained on Synthetic Data

(43:43) - Challenges faced & Career Advice

Our guest today is Harpreet Sahota, Deep Learning Developer Relations Manager at Deci AI. 

In our conversation, we first talk about Harpreet’s work as a Biostatistician and dive into A/B testing. We then talk about Deci AI and Neural Architecture Search (NAS): the algorithm used to build powerful deep learning models like YOLO-NAS. We finally dive into GenAI where Harpreet shares 7 prompting tips and explains how Retrieval Augmented Generation (RAG) works. 

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Link to Train in Data courses (use the code AISTORIES to get a 10% discount): https://www.trainindata.com/courses?affcode=1218302_5n7kraba

Follow Harpreet on LinkedIn: https://www.linkedin.com/in/harpreetsahota204/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

---
(00:00) - Intro

(02:34) - Harpreet's Journey into Data Science

(07:00) - A/B Testing 

(17:50) - DevRel at Deci AI

(26:25) - Deci AI:  Products and Services

(32:22) - Neural Architecture Search (NAS)

(36:58) - GenAI

(39:53) - Tools for Playing with LLMs

(42:56) - Mastering Prompt Engineering

(46:35) - Retrieval Augmented Generation (RAG)

(54:12) - Career Advice

Our guest today is Ryan Shannon, AI Investor at Radical Ventures, a world-known venture capital firm investing exclusively in AI. Radical's portfolio includes hot startups like Cohere, Covariant, V7 and many more. 

In our conversation, we talk about how to start an AI company & what makes a good founding team. Ryan also explains what he and Radical look for when investing and how they help their portfolio after the investment. We finally chat about some cool AI Startups like Twelve Labs and get Ryan’s predictions on hot startups in 2024.

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Link to Train in Data courses (use the code AISTORIES to get a 10% discount): https://www.trainindata.com/courses?affcode=1218302_5n7kraba

Follow Ryan on LinkedIn: https://www.linkedin.com/in/ryan-shannon-1b3a7884/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

---

(0:00) - Intro

(2:42) - Ryan's background and journey into AI investing

(11:15) -  Radical Ventures

(14:34) - How to keep up with AI breakthroughs? 

(22:42) - How Ryan finds and evaluates founders to invest in

(32:54) - What makes a good founding team? 

(38:57) - Ryan's role at Radical 

(45:53) - How to start an AI company 

(50:22) - Twelve Labs

(59:19) - Future of AI and hot startups in 2024

(1:09:48) - Career advice

Our guest today is Christoph Molnar, expert in Interpretable Machine Learning and book author. 

In our conversation, we dive into the field of Interpretable ML. Christoph explains the difference between post hoc and model agnostic approaches as well as global and local model agnostic methods. We dig into several interpretable ML techniques including permutation feature importance, SHAP and Lime. We also talk about the importance of interpretability and how it can help you build better models and impact businesses.

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Link to Train in Data courses (use the code AISTORIES to get a 10% discount): https://www.trainindata.com/courses?affcode=1218302_5n7kraba

Follow Christoph on LinkedIn: https://www.linkedin.com/in/christoph-molnar/

Check out the books he wrote here: https://christophmolnar.com/books/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

---

(00:00) - Introduction

(02:42) - Christoph's Journey into Data Science and AI

(07:23) - What is Interpretable ML? 

(18:57) - Global Model Agnostic Approaches

(24:20) - Practical Applications of Feature Importance

(28:37) - Local Model Agnostic Approaches

(31:17) - SHAP and LIME 

(40:20) - Advice for Implementing Interpretable Techniques

(43:47) - Modelling Mindsets 

(48:04) - Stats vs ML Mindsets

(51:17) -  Future Plans & Career Advice

Our guest today is Demetrios Brinkmann, Founder and CEO of the MLOps Community.

In our conversation, Demetrios first explains how he transitioned from being an English teacher to working in sales and then founding the MLOps community. He also talks about the role of MLOps in the ML lifecycle and shares a bunch of resources to level up your MLOps skills. We then dive into the hot topic of GenAI and LLMOps where Demetrios shares his view on specialised vs generalised LLMs and why it can be dangerous to build a startup on top of OpenAI.

Demetrios finally explains what the MLOps community is all about. They are organising live events in around 40 countries, a great podcast, a slack channel, some new courses on generative AI and much more. Check out there website here: https://mlops.community/

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Link to Train in Data courses (use the code AISTORIES to get a 10% discount): https://www.trainindata.com/courses?affcode=1218302_5n7kraba

Follow Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

----

(00:00) - Introduction

(01:50) - From English Teacher to MLOps

(08:32) - How to get into MLOps

(12:46) - MLOps and the ML Lifecycle

(22:54) - GenAI & LLMOps

(32:32) - Business Implications of Relying on OpenAI

(35:32) - The MLOps Community

(43:03) - Career Advice: The Power of Writing

Our guest today is Noah Gift, MLOps Leader and award winning book author. Noah has over 30 years of experience in the field and has taught to hundreds of thousands of students online.

In our conversation, we first talk about Noah's experience building data pipelines in the movie industry and his experience in the startup world. We then dive into MLOps. Noah highlights the importance of MLOps,  outlines the Software Engineering best practices that Data Scientists must learn and explains why we shouldn't always use Python. Noah finally shares his thoughts on the difference between MLOps and LLMOps, Python vs Rust and the future of the field.

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Link to Train in Data courses (use the code AISTORIES to get a 10% discount): https://www.trainindata.com/courses?affcode=1218302_5n7kraba

Follow Noah on LinkedIn: https://www.linkedin.com/in/noahgift/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

————
(00:00) - Intro

(02:14) - Building data pipelines in the film industry

(11:47) - Noah's experience in Startups 

(17:57) - What is MLOps? 

(20:52) - Why should Data Scientists learn Software Engineering?

(27:59) - Importance of MLOps

(30:54) - Rust vs Python

(43:48) - Why we shouldn't always use Python

(49:26) - Difference between LLMOps and MLOps

(53:50) - Security and ethical concerns with LLMOps

(56:27) - The future of the field

(01:08:41) - Career advice

Our guest today is Marianne Ducournau, Head of Data Science at Qonto and ex Data Scientist at Amazon and Uber.

In our conversation, we first discuss Marianne's first job in Data Science working in the public sector and managing a 10-15 people team. Marianne then talks about her experience at Uber and shares various projects that she worked on. We dive into price elasticity modelling and financial forecasting where her team built thousands of model to forecast financial metrics in multiple cities.  Marianne finally explains her current role as the Head of Data Science at Qonto and gives advice on how to progress in Big Techs and in your career.

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Link to Train in Data courses (use the code AISTORIES to get a 10% discount): https://www.trainindata.com/courses?affcode=1218302_5n7kraba

Follow Marianne on LinkedIn: https://www.linkedin.com/in/mborzic/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

————

(00:00) - Introduction
(02:12) - Marianne's Journey Into Data Science
(05:05) - Managing A 10-15 People Team In Her First Job
(10:02) - Pros And Cons Of Working In The Public Sector
(16:51) - Transition From The Public Sector To Uber
(22:25) - Price Elasticity Modelling
(35:42) - Building 1000+ Models For Financial Forecasting
(42:10) - Progressing In Big Techs
(45:01) - What Is Qonto And Marianne's Role There?
(48:08) - Understanding Qonto's Product
(49:29) - Building A Team As Head Of Data Science
(54:37) - Impact Estimation
(01:02:52) - Marianne's Advice For Career Progression

Our guest today is Christof Henkel, Senior Deep Learning Data Scientist at NVIDIA and world number 1 on Kaggle: a competitive machine learning platform.

In our conversation, we first discuss Christof's PhD in mathematics and talk about the importance of maths in a Data Science career. Christof then explains how he started on Kaggle and how he progressed on the platform to become the world number 1 amongst millions of users. We also dive into recent competitions that he won and the algorithms that he used. Christof finally gives many advice on how to win Kaggle competitions and progress in your career.

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Link to Train in Data courses (use the code AISTORIES to get a 10% discount): https://www.trainindata.com/courses?affcode=1218302_5n7kraba

Follow Christof on LinkedIn: https://www.linkedin.com/in/dr-christof-henkel-766a54ba/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

————

(00:00) - Introduction

(03:00) - How Christof Got Into The Field

(07:59) - The Role of Mathematics In Data Science Careers

(12:27) - Why Christof Joined Kaggle And How?

(21:11) - Reducing Model Overfitting 

(27:03) - Three Steps To Succeed On Kaggle

(33:56) - Kaggle VS Applied Machine Learning In Industry

(40:12) - How He Became World Number 1

(46:02) - A Recent Competition That He Won

(56:59) - His Role At NVIDIA 

(01:01:24) - Startup Experience 

(01:06:43) - Career Advice 

Our guest today is Davis Blalock, Research Scientist and first employee of Mosaic ML; a startup which got recently acquired by Databricks for an astonishing $1.3 billion.

In our conversation, we first talk about Davis' PhD at MIT and his research on making algorithms more efficient. Davis then explains how and why he joined Mosaic and shares the story behind the company. He dives into the product and how they evolved from focusing on deep learning algorithms to generative AI and large language models. 

If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.

Follow Davis on LinkedIn: https://www.linkedin.com/in/dblalock/

Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/  

————
(00:00) - Intro
(01:40) - How Davis entered the world of Data and AI?
(03:30) - Enhancing ML algorithms' efficiency
(12:50) - Importance of efficiency
(16:37) - Choosing MosaicML over starting his own startup
(25:30) - What is Mosaic ML? 
(37:34) - How did the rise of LLM aid MosaicML's growth?
(46:54) - $1.3 billion acquisition by Databricks
(48:52) - Learnings and failures from working in a startup
(01:00:05) - Career advice