In Case You Missed It in August 2025

ICYMI, today's episode provides the top moments from conversations I had with my podcast guests in August. Interestingly, in August all of these episodes took off on YouTube in an unprecedented way:

1. Julien Launay, co-founder and CEO of the wildly successful Adaptive ML, explains how LLMs are trained including the pre-training and the increasingly critical post-training phases.

2. Michelle Yi provides her brilliant take on shocking misalignment research from Anthropic, which showed that A.I. agents will frequently resort to blackmailing humans when its goals are threatened.

3. Kirill Eremenko, CEO of the learning platform SuperDataScience, describes what it takes to become an A.I. Engineer.

4. And Akshay Agrawal who explains how his rapidly-adopted marimo notebook can be converted into a full-blown, click-and-point web app in *seconds*.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Hopes and Fears of AGI, with All-Time Bestselling ML Author Aurélien Géron

Aurélien Géron is the best-selling author of ML books of all time. Today's episode is his first interview in a *decade*... and it's a stunner! He reveals his next book as well as his deep thoughts on AGI.

Aurélien:

• Is the author of O'Reilly's "Hands-On Machine Learning" series of books. The fourth edition in the series will be on bookshelves in the coming months.

• Was Co-Founder and CTO of Wifirst, an exceptionally successful French tech company.

• Was previously a product manager at Google.

• Holds an MEng in Computer Science from AgroParisTech.

The start of today's episode (on Aurélien's books) may appeal primarily to hands-on practitioners like data scientists and software developers. The bulk of the episode, however, will appeal to anyone looking to understand how Artificial General Intelligence (AGI) will overhaul work, life and society for humans.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

8 Steps to Becoming an AI Engineer, with Kirill Eremenko

The #1 fastest-growing role in many countries (including the US) is A.I. Engineer. Want to become one? ...or add the skillset into your existing role? Today, Kirill Eremenko provides an 8-step roadmap!

Many of you will already know Kirill:

• Founder and CEO of SuperDataScience.com, the eponymous e-learning platform.

• Founded the SuperDataScience Podcast nine years ago and hosted the show until he passed me the reins five years ago.

• With over 3 million students, he’s the most popular data science and A.I. instructor on Udemy.

• He holds a Master’s from The University of Queensland in Australia and a Bachelor’s in Applied Physics and Mathematics from the Moscow Institute of Physics and Technology.

Today's episode will be of primary interest to hands-on practitioners like data scientists and software developers.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

How to Jailbreak LLMs (and How to Prevent It), with Michelle Yi

Today, extraordinary Michelle Yi details LLM jailbreaking (as well as data poisoning, prompt stealing and slop squatting!) and how to prevent it. Scary content but she makes it funny and entertaining, enjoy!

When I say "extraordinary", I'm not exaggerating. Michelle:

• Finished her undergrad at the same age as most folks finish high school.

• While working full-time as an engineering lead at IBM on Jeopardy-playing Watson, she was also a professional violinist in the New York Philharmonic!

• In the past decade, has held a impressive list of AI leadership roles at Bay Area startups.

• Now is helping (startlingly underrepresented) women in tech startups and venture capital through co-founding Generationship, being a venture partner in (ironically named) The Tech Bros and a board member for Women In Data™️.

Today's episode skews a bit toward hands-on practitioners but Michelle does such a wonderful job of communicating complex concepts and making them relevant to modern global events that anyone might love this episode.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Data Lakes 101 (and Why They’re Key for AI Models), with Oz Katz

Today, Oz Katz joins me to explain what data lakes are as well as what data infrastructure's needed to train and run modern A.I. models (such as multi-modal LLMs). This is an informative one!

Oz Katz:

• Is co-founder and CTO of lakeFS, which looks and feels like Git but is for data (instead of software) versioning and collaboration.

• Was previously co-founder/CTO at a number of other tech startups, including Swayy, which was acquired by Similarweb in 2015.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

LLM Pre-Training and Post-Training 101, with Julien Launay

How are cutting-edge LLMs are trained? Find out in today's exceptional episode with Julien Launay, who digs into pre-training (supervised learning) and post-training (reinforcement learning) in eloquent detail.

Julien:

• CEO and co-founder of Adaptive ML, a remarkably fast-growing startup focused on enabling A.I. models to learn from experience.

• Previous led the extreme-scale research teams at Hugging Face and LightOn, where he helped develop state-of-the-art open-source models.

• Organizer of the "Efficient Systems for Foundation Models" workshop at ICML (the prestigious International Conference on Machine Learning).

Today's episode will appeal most to hands-on practitioners but other folks who are open to getting into the technical weeds on Large Language Model (LLM) training should also listen in.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

In Case You Missed It in July 2025

Wow, what a month... ICYMI, here are the top moments from the conversations I had with my podcast guests in July:

1. AI-lab director Lilith Bat-Leah on why data-centric machine learning research (DMLR) is the future of AI.

2. Prolific author and speaker Sinan Ozdemir (last week in San Francisco, I witnessed him deliver the *best* talk I've ever seen) on whether we can trust LLM benchmarks.

3. Bloomberg's Dr. Sebastian Gehrmann on why generic LLMs fail in regulated industries like finance and healthcare.

4. AI entrepreneur Dr. Zohar Bronfman on how AI can predict what you'll do next... before you're even consciously aware of your decision.

5. Microsoft AI researcher Dr. Robert Osazuwa Ness with step-by-step guidance on how to build causal AI models using PyTorch.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

The Future of Python Notebooks is Here, with Marimo’s Dr. Akshay Agrawal

I love Jupyter Notebooks... but they have a lot of painful "features". Today's guest Akshay Agrawal has built marimo, which resolves these issues and adds in lots of clever new innovations.

More on Akshay:

• Co-founder and CEO of marimo.

• Carried out a PhD in electrical engineering at Stanford University.

• Previously held software engineering roles at Google and Netflix.

Today's episode will appeal most to hands-on practitioners. In it, we cover:

• Why 96% of Jupyter notebooks fail to reproduce their original results.

• How reactive notebooks can transform a simple slider adjustment into automatic recalculation across your entire analysis — like Excel for data science.

• How you can now intuitively select data points with your mouse in a scatter plot and instantly get them back as a Python dataframe for analysis.

• How one marimo notebook can simultaneously be an executable script, an importable Python module, and even a fully functional click-and-point UI!

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Causal AI, with Dr. Robert Usazuwa Ness

Today's guest, Dr. Robert Osazuwa Ness, wrote the popular new book "Causal A.I." so enjoy this episode on what Causal A.I. is and what advantages it has over "normal" (correlation-based) models.

Robert:

• Senior Researcher at "Microsoft Research A.I."

• His research focuses on statistical and causal inference techniques for controllable, human-aligned multimodal models.

• He is also founder of Altdeep.ai, where he teaches professionals advanced topics in machine learning.

• Holds a PhD in Statistics from Purdue University in Indiana.

Today’s episode will resonate most with hands-on practitioners like data scientists, statisticians and A.I. engineers.

In today’s episode, Robert details:

• The three-rung ladder of causation that determines what types of causal questions you can actually answer with your data.

• The surprising connections between Bayesian networks, graphical models and modern causal A.I.

• Why A.I. systems have been dominated by correlation-based learning and what's stopping them from adopting causal reasoning like humans and animals naturally do.

• How tools like PyTorch, Pyro, and DoWhy are revolutionizing causal inference by separating statistical complexity from causal assumptions.

• How large language models like GPT-4o can act as "causal knowledge bases" and outperform traditional causal methods in some scenarios.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Guest Co-Host of Last Week in A.I., Episode #217

The *only* podcast I listen to is "Last Week in A.I." — and, for this week's episode, I had the privilege of co-hosting it! Get caught up on all the biggest A.I.-related news of the past week, including:

• OpenAI's new "ChatGPT Agent", which can control computers and perform a wide range of tasks, showcasing powerful performance benchmarks and potential applications in business and research.

• xAI's racy new romantic chat... complete with stripping down to lingerie.

• Major business moves in the A.I. space, including Google's acquisition of Windsurf's top talent after OpenAI's deal fell through, and several notable hires by Meta from OpenAI and Apple.

Thanks to Andrey Kurenkov and Jeremie Harris for doing such a tremendous job of curating and hosting this weekly news show — as well as for welcoming me back to co-host again :)

This week's episode of "Last Week in AI" is #217. You can catch it on your favorite podcasting app or on YouTube.

Neuroscience, AI and the Limitations of LLMs, with Dr. Zohar Bronfman

I was blown away by today's guest, the brilliant dual-PhD Zohar Bronfman as we discussed neuroscience, A.I., and why predictive models offer a better ROI than generative ones. Enjoy!

Dr. Bronfman:

• Is the co-founder and CEO of Pecan AI, a predictive analytics platform that has raised over $100m in venture capital.

• Holds two PhDs — one in computational neuroscience and another in philosophy — bringing a deep, multidisciplinary lens to the design and impact of A.I. systems.

• Focuses on the evolution of machine learning from statistical models to agentic systems that influence real-world outcomes.

Today’s episode will be fascinating for every listener.

In it, Zohar details:

• The trippy implications of the reality that your brain makes decisions hundreds of milliseconds before you're consciously aware of them.

• The intelligence feat that bumblebees can do that current A.I. cannot, with implications for the realization of human-like intelligence in machines.

• Why predictive models are more important than generative models for businesses but how generative LLMs can nevertheless make building and deploying predictive models much easier and accessible.

• The rollercoaster journey that led him to create a sensationally successful A.I. startup immediately upon finishing his academic degrees.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

How Prof. Jason Corso Solved Computer Vision’s Data Problem

Exceptional episode for you today with Prof. Jason Corso, in which he details how he's tackling the biggest problem in machine vision. Jason's super sharp and very well-spoken... don't miss this one!

Jason:

• Professor of Robotics, Electrical Engineering and Computer Science at the prestigious University of Michigan, with over 20 years of research spanning video understanding, robotics, and AI.

• Has published over 150 academic papers that, together, have been cited over 20,000 times.

• Co-founder and Chief Science Officer at Voxel51, a leading platform for visual AI development.

• His work bridges academic innovation and real-world impact, earning him more major honors than I have space to list!

Today’s episode skews a bit toward hands-on practitioners like data scientists and AI/ML engineers, particularly anyone tackling computer-vision problems. That said, Jason is a charismatic and exceptional communicator so perhaps any listener to this podcast will enjoy today’s episode.

In it, Jason details:

• How his research spinout, Voxel51, is solving the biggest bottleneck in computer vision.

• The surprising way autonomous vehicles learn to handle accidents they've never seen.

• Why the secret to better AI models isn't better algorithms — it's something else that’s hiding in plain sight.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

Why RAG Makes LLMs Less Safe (And How to Fix It), with Bloomberg’s Dr. Sebastian Gehrmann

In today's episode, A.I. researcher Dr. Sebastian Gehrmann details what RAG is and why it makes LLMs *less* safe... despite popular perception of the opposite.

Sebastian:

• Is Head of Responsible A.I. at Bloomberg, the New York-based financial, software, data, and media company that (with 20,000 employees) is huge.

• Previously, as Head of NLP at Bloomberg, he directed the development and adoption of language technology to bring the best A.I.-enhanced products to the Bloomberg Terminal.

• Prior to Bloomberg, was a senior researcher at Google, where he worked on the development of large language models, including the groundbreaking BLOOM and PaLM models.

• He holds a Ph.D. in computer science from Harvard University.

Today’s episode skews slightly toward our more technical listeners like data scientists, A.I. engineers and software developers, but anyone who’d like to be up to date on the latest A.I. research may want to give it a listen.

In today’s episode, Sebastian details:

• The shocking discovery that retrieval augmented generation (RAG) actually makes LLMs LESS safe, despite the popular perception of the opposite.

• Why the difference between 'helpful' and 'harmless' A.I. matters more than you may think.

• The hidden “attack surfaces” that emerge when you combine RAG with enterprise data.

• The problems that can happen when you push LLMs beyond their intended context window limits.

• What you can do to ensure your LLMs are Helpful, Honest and Harmless for your particular use cases.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

LLM Benchmarks Are Lying to You (And What to Do Instead), with Sinan Ozdemir

Sensational episode for you today with the illustrious A.I. author, educator and entrepreneur Sinan Ozdemir on how LLM benchmarks are lying to you... and what you can do about it.

Sinan:

• Is Founder and CTO of LoopGenius, a generative A.I. startup.

• Authored several excellent books, including, most recently, the bestselling "Quick Start Guide to Large Language Models".

• Hosts the "Practically Intelligent" podcast.

• Was previously adjunct faculty at The Johns Hopkins University, now teaches several times a month within the O'Reilly platform.

• Serial A.I. entrepreneur, including founding a Y Combinator-backed generative A.I. startup way back in 2015 that was later acquired.

• Holds a Master’s in Pure Math from Johns Hopkins.

Today’s episode skews slightly toward our more technical listeners but Sinan excels at explaining complex concepts in a clear way so today’s episode may appeal to any listener of this podcast.

In today’s episode, Sinan details:

Why the A.I. benchmarks everyone relies on might be lying to you.

• How the leading A.I. labs are gaming the benchmark system.

• Tricks to actually effectively evaluate LLMs’ capabilities for your use cases.

• What the future of benchmarking will involve, including how to benchmark agentic and multimodal models.

• How a simple question about watermelon seeds reveals the 40% failure rate of even today’s most advanced A.I. models.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

In Case You Missed It in June 2025

We had exceptional guests on my podcast in June. In today's "In Case You Missed It" episode, hear the best parts of all my June convos. Here's a quick summary to tantalize you:

1. Strategy consultant Diane Hare with five tricks for gaining buy-in on A.I. transformation in your organization.

2. Renowned data-career educator Avery Smith on the two portfolio projects every aspiring data analyst should build.

3. SuperDataScience founder Kirill Eremenko on what you're missing if you're struggling to land an A.I. job.

4. San Fran-based venture capitalist Shaun Johnson on the traits that make a great A.I. startup founder.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.