Matt Harrison

Where are the Pandas, Polars, XGBoost reviews/speedups on this thing?

Thanks for sharing your experience. Sad that the interface is to "use Docker"...

I like the Aider leaderboard because the cost is not reported directly in terms of tokens, but rather in terms of the total amount to solve the problems. With thinking tokens these days, who knows if the actual cost is 10x the advertised cost? (Note that the cost is on a logarithmic scale)

This is the chart that I'm paying attention to. (Of course, it is outdated very quickly these days.)

There is an "efficient frontier" where you get the most bang for your buck.

Certification Programs Showcase your skills and advance your career by getting NVIDIA-certified.

Each certification comes with a curated set of recommended free and paid training to help you prepare, so you can build confidence and prove your expertise.

nvda.ws/4mDFSB4

AI isn’t going anywhere, and neither should your skills.

NVIDIA already offers certifications in Generative AI and LLMs & Multimodal Generative AI, and now they are raising the bar with two new professional-level certifications: Agentic AI and Generative AI LLMs.

Expensive but getting closer

Sign up for the Oct 7th webinar to learn more and get ahead of the curve. During the webinar, you’ll get preparation tips, try sample exam questions, and as a bonus, receive a special exam promo code.

nvda.ws/46CVW01

Each certification comes with a curated set of recommended free and paid training to help you prepare, so you can build confidence and prove your expertise.

AI isn’t going anywhere, and neither should your skills.

Do you know where the gaps are in your AI knowledge?

NVIDIA already offers certifications in Generative AI and LLMs & Multimodal Generative AI, and now they are raising the bar with two new certifications: Agentic AI and Generative AI LLMs.

Love these charts showing the efficiency boundaries...

Spent the day getting local models running on my laptop and hooking up Aider to it.

(Some might find it interesting that AI was useless in helping me do this. Needed to use old-fashioned sleuthing skills.)

(Nor is this documented anywhere that I could find.)

Funny that the pelican looks like it crashed

It’s a strange moment as an author. On one hand, validating that our work has real value in this new AI economy. On the other, frustrating that consent wasn’t part of the process.

Let's see what happens and how much of this $60k I actually receive...

By my calculation, @anthropic.com owes me around $60,000.

There was a recent settlement where Anthropic is paying authors $3,000 per book that they illegally acquired and used for their training data.

I've written a few books and can find almost 20 of mine in their stolen content.

Loved adding "Ancombe's Quintet" to my curve fitting exercise in my class this week.

Thanks Carl McBride Ellis, PhD

I'm teaching two Pandas courses this week.

Here's a "whiteboard" from one of them.

Looks like I need to correct it now as @pola.rs spans from small to big data now!

Congrats to Ritchie Vink and team!

This post caused me to read the candidate statements. Though I don't think Franz could ever win, I do find his ideas interesting (and a change would be somewhat refreshing, though perhaps not this). And he was one of the few candidates who had concrete goals. Most of them were very ambiguous.

Whoops. I just did an online pandas training...

Did I say that?

Really busy teaching recently.

This week I'm teaching two Pandas/data analytics courses.

Here's a drawing I made during last week's ML course.

Here's a drawing I made during my course yesterday to help students remember the difference between precision and recall. What do you think of it?

(As always, follow me for more insights on Python, tabular data science, and AI applications. And if your team needs help with these tools, reach out.)

“Can’t AI just solve ML problems for me?”

Again, sometimes. But without an understanding of the fundamentals, you can’t sanity-check the output. And trust me, AI goes off the rails more often than you’d think. (Ask me how I know…)

Now, I often get asked: “Can’t AI just replace ML?”

The honest answer: sometimes. But it’s usually more expensive, slower, and often delivers worse performance.

I'm teaching a machine learning class this week.

One of my favorite parts of teaching is helping technical folks not just learn the concepts, but also see how they apply directly to their own domains.

What do you think of my newest shirt?

YouTube video by Matt Harrison How I Use an Insta360 Link with Teleprompter

In this video, I discuss lighting, cameras, (Insta360) webcams, and teleprompters.

Images below show my recreation of 90% of folks' camera setup and mine.

Let me know if you have any questions.

youtu.be/IG-H5poJMvY

One unique feature of my office is a teleprompter. I use it for:

Looking in the eyes during Zoom calls
Live teaching
Recording courses

My primary goal is to present myself professionally and provide my students with the best possible experience.

Folks often ask me what I do for camera/lighting.

I'm not a videographer, lighting, or sound specialist. Yet, I've been in a couple of studios and cherry-picked ideas that minimize the drag on my environment, allowing me to complete the work I need to do as efficiently as possible.

Optimize your environment.

I've spent a lot of time optimizing my environment for the type of work I do. "Your video quality and background look great" is a common piece of unsolicited feedback.