I step out of the apartment into 63 degree sunshine. A light breeze is stirring.

To my left is the tiny cafe I had sopa (soup) in a few days ago.

After a short block and a right turn I’m saying hi to the barber at St. Patrick’s Barber Shop who cut my hair from my badly translated instructions a few days before. While my Spanish was bad, the cut was great.

He’s helping the shopkeeper lady next door get her clothing racks organized on the sidewalk.

I pull off my baseball cap and, grinning, point to my mussed hair, “Bien, bien,” I say, nearly exhausting my Spanish vocabulary (not really but close enough).

Two blocks down a man, in his seventies maybe, is hawking calendars. He doesn’t have a shop, he’s just walking the streets with maybe three or four unmatched, spiral bound 2026 calendars.

I approach. “Quince,” he says after a handful of other words I don’t understand. He’s saying fifteen thousand Colombian Pesos, which is about $3.75 USD. I hand him a 20,000 and tell him to keep the change. It takes a few back-and-forths before he realizes I’m telling him to keep it.

Everything here is cheap by US standards. Lunch $5, dinner $6-$7, bottled water $0.75, beer $0.75, my AirBNB (I went bargain basement) $35/day. Fifty per day would have been luxe.

A few blocks later is Café Revolución, the place where proprietor, Zsolt (pronounced “zholt”) also operates an animal sanctuary. The ranch is about 250 miles to the southeast of here and tends to a variety of animals like crocodiles and capybara, and other species.

(The cafe name reminds me of this perfect segment in the movie The Trip, with Steve Coogan and Rob Brydon. “We rise at 9:30.”)

Two days later, Zsolt says, “Hey hang on I have something for you.” He hands me the last of his “Capy Project” stickers to give to my daughter back home. See more about the Capy Project.

Paradise?

My digital nomad friends say, “the weather,” when I ask what they like about Medellín. See the nearby table. Now that is some steady perfection if you ask me.

Most mornings are cool with a light breeze and afternoons are warm but not hot, certainly not by my hometown standards (Austin, can you say sixty days over 100?).

Flora and Fauna (er, Canines)

One thing about the dogs here is that so many are off-lease - everywhere. On the sidewalk, in the parks, crossing busy streets. This is new to me.

These pups are not racing around, getting into trouble. No, they walk the sidewalks just like people.

Someone told me they are just socialized into normal life, sort of like children I guess. And it makes sense. If you raise a pup to go everywhere with you, he learns cues from you, the surroundings, traffic, and learns to navigate without bolting into the street.

Yesterday, I saw a man and his golden retriever crossing at a crosswalk. When the man sped up to make the light the dog did too.

I discovered this handsome trio at the riverside Christmas light display. Note the wrestler-style gold chain complete with heavy medallion one is wearing. Clever styling for sure. Hit them up on the socials at @leonriko319.

Flora

Laureles, the neighborhood I stayed in, is a riot of gorgeous greenery and flowers.

Enjoy a bit of this urban beauty. See many more photos at the end of the post.

The Streets of Laureles

Admittedly, my experience of Medellín is limited. I’ve mainly stayed within one square mile in the Laureles neighborhood, walking from my AirBNB to a co-working space, shops and restaurants.

The city does have an edge, certainly to someone from the suburbs of Austin, TX. Don’t get me wrong, I’ve seen plenty of seedy places, having lived in NYC, Brooklyn, New Orleans and Houston. (I was once mugged just off Bourbon Street when I was sixteen. I know, sixteen, sheesh.)

Can one know a place after just two weeks? No, not really.

Last night as I rounded the corner to make for my apartment door, I saw a man remove his last bits of clothes and, entirely naked, squat down and defecate on the sidewalk twenty feet from my doorstep.

Taxis and Trash Pickup

Ok so you know the taxi routine. Walk out of the airport, stand in a long line. When it’s your turn, step into the next taxi at the front of the line.

Colombia? Nope. Other way round.

I witnessed this humorous bit of cross-culture difference outside an Éxito grocery store at Unicentral Mall. Rather than step to the front of the line of taxis, Colombians board at the back. And off you go.

Then, the magic starts.

Each driver, standing outside his car, releases the parking brake and pushes his taxi, gently rolling it one spot backward. He’s standing outside the car. Amazing. And gas-saving I guess?

At 7:10 this morning I watch the garbage collection version of cultural difference. See, unlike the American trash collection I know, on my street this morning the garbage men, clad in neon Day-Glo orange of course, ring a bell to politely let everyone know it’s time to bring out the trash. Ingenious.

Not only is this a quirky communal experience, it also solves the unfortunate problem that homeless folk will tear open plastic garbage bags to get at your leftovers, leaving a big mess.

More Plants and Flowers (many more)

Almost all of these plants I found around the Laureles neighborhood, bounded by the triangle formed by Avenida Nutibara, Avenida San Juan and Carerra 66B.

When I experienced sailing on a family trip in the British Virgin Islands in 1995, sailing gripped hold of me and never let go. It only tightened its grasp on me since then.

There is something about the blend of man’s engineering with wind and water that stirs my soul. Add in the history of the 174 year-old Cup and it’s a marvelous thing to behold.

So, when last year I found out that the America’s Cup yacht race publishes telemetry on matches, I jumped at the chance to get my hands on the data.

Racing Data

After preliminary races in places around the world, the 2024 America’s Cup finals started in Barcelona, Spain. I found a directory of racing data at the America’s Cup website available for public download.

Using Python (a popular programming language in data analytics), I was able to use this public data for some fun projects. I calculated boat speed, location and direction for each boat around the entire 25 minute races. Using this I also plotted performance graphs and charts showing the paths and turns of each competitor. See image nearby of the courses shown on a map of the Barcelona waterfront.

Data Cleaning and Preparation

The vast dataset was tricky to get started with. The race committee did not provide a key to the JSON (a computer format) files. The files are chock full of course data, like boundary locations, GPS coordinates, Unix timestamps, locations of floating marks and sundry other race data. What I wanted was boat specific data like speed and location. After some meticulous trial and error I was able to figure out the boat IDs, timestamps, GPS locations and more.

The Possibilities

With so much rich data on boat performance, this raises all sorts of potential analyses. You could measure boat speeds over different parts of the course to see which teams took advantage of “fast sides” of the course.

You could analyze tacking performance by calculating how much velocity is lost when each boat tacks. Or you could measure average velocity made good (VMG), an all-important measure among yacht racers.

Next Stop Italy

The 38th America’s Cup is planned for 2027 in Naples, Italy. Can’t wait to follow it and I’m sure the technical and data elements will be even more impressive.

AI continues to change at an amazing rate. Trying to keep up is tough even for the insider-experts. I originally published this list of my favorite AI resources in December 2024. Since then, I’ve discovered new gems worth sharing, and I’ve updated this post to reflect what I’m using and learning from today.

These podcasts, newsletters, tools, and courses help me stay sharp as I grow my skills in AI and ML.

(new) - means I added this since my December post.

Podcasts

(new) Dwarkesh Patel - Amazing wunderkind (24 years old as I write this) podcaster. Has interviewed serious heavy hitters, including Mark Zuckerberg, Tony Blair (yes, that Tony Blair), Satya Nadella (CEO Microsoft). Not dedicated to AI but touches often.

(new) Lex Fridman - Another amazing podcaster. Super smart interviewer, scientist and Jiu-Jitsu practitioner. Unique in the world of podcasting: goes very long and deep (episode with Elon and the Neuralink team was 8.5 hrs + !!), quiet calm voice, very thoughtful. Not dedicated to AI, but touches often.

Last Week In AI – founded by a Stanford grad with a PhD in AI. Hosted by super smart
duo, Andrey Kurenkov and Jeremie Harris, who share headlines and deeper stories.

The AI Daily Brief – quick daily hits. Interesting and informative.

The Super Data Science Podcast – more focused on the field of data science but also has some great interviews, for instance recently with Andrew Ng of Deepmind.AI and Coursera. Fantastic host Jon Krohn.

Newsletters

(new) Ethan Mollick - respected prof. at the Wharton School and AI expert. I’m starting to hear and read more and more from Mollick.

(new) Ethan Holland - I’m just getting acquainted with this one. A HUGE assortment of news every week. MUST be AI collated (?) given the breadth of info.

Don’t Worry About the Vase by the incredible Zvi Mowshowitz, a fascinating and prolific writer. Economist and Magic: The Gathering gaming pro (as in played as a professional). ** NOTE: you can also find this in ‘podcast’ form. It’s not really a podcast, but an audio version of the written blog voiced by an ElevenLabs bot.

Top AI Chat Models (LLMs)

(new) Meta AI - Meta’s (formerly Facebook) research group building cutting-edge AI models like Llama, with a big focus on open science and sharing their work with the world.

(new) Deepseek - New model out of China; made waves in early 2025 due to its high performance at a lost cost/price point.

(new) Kimi K2 - new model out July 2025. High test results so far, known for good creative writing.

ChatGPT – the big guys on the block, top chat LLM. Part of OpenAI offerings, with large investments from Microsoft. (I have a paid subscription)

Claude – Advanced LLM by Anthropic, a company started by OpenAI alum and focused on safety/alignment. Very good with code. (I have a paid subscription)

Perplexity – A powerful model which places reference links for all of its output, emphasizing sourcing for its LLM responses.

XAI – Elon Musk’s AI from X. Interesting and sometimes zany. I’m waiting for the “Unhinged” model to come out soon. Really, Google it.

Gemini – Google’s frontier model. Ver. 2.5 Pro is really impressive. If Google is going to stay relevant in the new, fast moving world of AI this is it.

NotebookLM – a unique take on LLMs. You can feed in up to 50 large documents and even YouTube videos and ask questions about the data contained therein.

AI Tools (new)

I’ve added a new section in July 2025 for tools I find interesting or must-have.

(new) Gamma - for making slides easy and fast.

(new) Higgsfield - for making still images into short AI video.

Education & Courses

(new) - Alpha School. This is not for AI education per se, but using AI in “traditional” education settings. And it’s local to Austin, where I live. Great piece by Zvi on Alpha School. Raises so many interesting ideas and questions. Read a parent’s in-depth review of the school here. (I have not read this yet, it’s 18,000 words!)

(new) Huggingface – Learn and share AI tools and models. Like GitHub for AI, it's where tons of cool models, datasets, and tools live, and anyone can use or share them.

(new) Kaggle.com - large community for learning and AI contests, some with huge cash prizes. I competed in one of the competitions (and wrote about it).

(new) Agents - free 4-hour technical course from Ed Donner and Jon Krohn on how to build agents. Get ready for some Cursor, Python and API work. Highly recommend.

DeepLearning.AI (Coursera) – deeplearning.ai was founded by Andrew Ng, one of the pioneers in AI and machine learning education. Many of the courses, like the highly-regarded Deep Learning Specialization, are hosted on Coursera.

University of Texas – McCombs School of Business – Post-Grad Program in AI and Machine Learning – the seven month course I took through UT (finished April 2025). Highly recommend. I thoroughly enjoyed this course and learned a ton of fascinating and practical things. The mathematics of AI and how LLMs and transformers work blew my mind.

Marc Andreesen: “I think there's a plausible argument—which Elon also believes—that robotics is going to be the biggest industry in the history of the planet. It's going to be gigantic. There will be billions, tens of billions, even hundreds of billions of robots of all shapes, sizes, and descriptions running around doing all kinds of things.” — (From the Reagan Foundation Economic Forum Simi Valley, California May 30, 2025)

Catch that? Billions of robots doing all sorts of things.

As the recently released report on AI from Mary Meeker says, things really are moving faster than we realize. Advances in AI are begetting ever faster development of the next big thing.

Robotics has advanced a ton in recent years, so I wanted to update my knowledge on what’s out there. Join me on a short tour of some of the main players.

Humanoid Robots

Some background.

The world of robotics is now enormous. Not in the billions yet, but still big. The International Federation of Robotics claims 3.9 million robots were in use worldwide by 2022. That is all robot types not just the human-like robots we see from Tesla and Apptronik (these humanoid bots are a very small part of the pie). Much more common are the robotic arms we see in auto and electronics manufacturing.

Let’s take a look at some of the leading players in the world of human-like, humanoid robots. Because of the immense complexity of a humanoid, versus say a single-purpose, fixed industrial robot, this field is still just getting off the ground.

The Big Players

Some of the top makers of human-like robots include Tesla, Apptronik, Sanctuary AI, Figure AI, Agility Robotics and Boston Dynamics.

Tesla

Tesla made a big splash in October 2024 with the “We Robot” event in Los Angeles. The company unleashed both a robotaxi and the humanoid Optimus robot at the same time. (By the way, for my local readers, Tesla plans to premier the robotaxi here in Austin in June 2025.)

Optimus is geared to “…performing unsafe, repetitive or boring tasks,” according to Tesla. I remember the buzz after the bot was introduced in the October event. People were gushing about how it was making drinks for them and talking with them.

Tesla standout feature: Learning by watching people. According to Milan Kovac, VP of Optimus, the bot learns from third-person videos on the internet of people doing tasks.

Note that much of what Optimus can currently do requires human operators (tele-operators) controlling things from afar. But the goal is to get to fully-autonomous operations without constant human intervention.

Given Elon’s enormous capacity to lead word changing technologies, I do have high hopes for what’s coming from Tesla.

Apptronik

In August 2023, Apptronik unveiled its Apollo robot. At roughly human size, 5 feet 8 inches tall and 160 pounds, the Apollo looks innocently human-like. It can lift and carry 55 pounds.

“People don’t want to do robotic, physically demanding work in tough conditions and they shouldn’t have to,” according to Apptronik.

The company is a spin-off of work started at The University of Texas and now has funding from Google among others. Among the bots I reviewed, Apollo appears the most human-like with eyes, head shape, arms and legs that are all vaguely human.

Apptronik Standout feature: Option to mount permanently to the floor with a static base (no legs), retaining the upper torso as-is.

See images of the Apptronik Apollo bot here.

Figure AI

This robot has figured (sorry, couldn’t resist) prominently alongside automaker BMW’s content and stories. It’s debatable how much the robots actually do at the BMW plant, but something is definitely happening.

Figure's robots will enable companies to increase productivity, reduce costs, and create a safer and more consistent environment," said Brett Adcock, Founder and CEO of Figure. "We look forward to working side-by-side with BMW Manufacturing to integrate AI and robotics into automotive production," said Adcock.

According to Forbes, some number of the robots are working at BMW but it seems the real life usage is limited to only a few bots and mostly at off hours.

On Figure’s website, they target three core use cases: physical labor, household and off-world! Now that’s my kind of forward thinking. Not even limited to this planet.

Figure standout feature: Collaboration with BMW manufacturing to use robots on the factory floor.

Sanctuary AI

According to the company, “Sanctuary AI is on a mission to create and deploy millions of industrial-grade humanoid robots to help address growing labor challenges.”

Of note, Sanctuary had an investment in another manufacturer on this list. In April Sanctuary announced it is selling its stake in Apptronik, the Texas robot maker, worth $125 million, according to The Logic.

Sanctuary standout feature: Hands. According to Sanctuary, their bot has “Industry-leading robotic hands with increased degrees of freedom (20 in total) that rival human hand dexterity and fine manipulation with proprietary haptic technology that mimics the sense of touch…” This should go a long way toward bringing robots the fine motor skills that humans have.

This video from Sanctuary blew me away. In it they describe something I had not thought about: we humans can’t always see everything we touch and interact with. Think of rummaging in your pocket for keys or digging into a drawer to grab a pen. We don’t necessarily see everything, we navigate by touch instead. This touch capability capability is one that Sanctuary is focused on.

Agility Robotics

“Meet DIGIT. The world’s first commercially deployed humanoid robot,” says Agility on their website. None other than Amazon has been testing Digit for warehouse operations, according to the retail and cloud giant. Amazon also has an investment stake in Agility through its Amazon Industrial Innovation Fund.

As far as I can tell, Agility has the only ready-made list of “skills” to choose from on their website. These include skills like tote palletizing, tote stacking and conveyor unloading.

The company discusses “Robots as a Service,” riffing on the ubiquitous SaaS models of software sales.

Agility standout feature: Amazon involvement. Given the enormous capability and innovative reputation of Amazon, I’d take note of Agility and expect it to have long legs.

Boston Dynamics

Boston Dynamics is an early, impressive innovator. They’ve been around since 1992 when Marc Raibert spun off from the MIT Leg Lab to form Boston Dynamics. Fast forward to 2025 and the company is now owned by Hyundai Motor Group and offers an array of robotic automation.

The Atlas humanoid robot is just one model from Boston Dynamics. Spot is popular as you’ve likely seen the bright yellow dog-like robot before. The company also sells more common fixed industrial robot arms as well.

Boston Dynamics standout feature: Long string of successful innovation over decades. Full range of commercially viable robots. Given their track record I have high hopes for Boston Dynamics.

Here is a summary of specifications for the robots we’ve looked at.

What’s Next

The holy grail of robotics is autonomy - the ability of a machine to proceed self-directed in hours (or more) of activity without ongoing human direction. Imagine a home domestic bot that starts in the morning and does all of its daily chores without you having to tell it what to do next.

We are still on the cusp of autonomous operation. Each of the companies offering humanoid bots is still limited, on the one hand by requiring tele-operators and on the other, severely limited to controlled, smaller areas of operation with limited freedom of movement.

A common theme among the six bots I covered is an emphasis on warehouse and logistics use cases. These workloads involve repetitive, potentially boring, dangerous tasks. What better place to start helping human operators. Elon loves to tout his bot in homey settings like the family kitchen, but those times are probably further out than the industrial applications.

Until now.

Hailing the Car of the Future

In Tempe/Phoenix you can download the dedicated Waymo app and summon one of the fleet of specially equipped Jaguar SUVs. I pulled up the app while we were saying our goodbyes in the parking lot of a restaurant. I mention this detail because the pickup spot is super accurate. When our taxi arrived it went to that exact spot. Later, I did see I could have adjusted the pickup spot but this was completely fine with us.

When the car arrived the app instructions on my phone were pretty clear, although it was slow to unlock the doors. We pressed the button on the left passenger door several times and it didn’t open at first. Additionally, I tried to put our luggage in the trunk but couldn’t figure out how to get it open. So instead I squeezed our roller bag from the backseat into the front - a bit awkward. As I was shoving the bag forward I wondered if there was sensor that would catch me and yell out, “stop.” But thankfully it was fine.

And We’re Off!

The departure was smooth and pleasant. A rear-facing screen mounted to the center console welcomed us and explained a few safety related items in both video and voiceover. The screen had controls for ac temp, sound volume and music choice. It also kept us updated on our progress to the airport. It was a busy Saturday afternoon and so the five mile trip took about 20 minutes, on moderate speed surface streets.

The first turn was a bit unsettling.

We’ve all been there. The tricky left across multiple lanes. Our car chose a route going left on a busy, divided road (two lanes each way plus a center turn lane). Since traffic was heavy, cars were backed up and blocking our access to the center turn lane. So, cars are coming from our left and the lane we need to be in is full. What to do? This gave us an immediate example of how humans and machines interact differently.

A driver in our intended lane saw us struggling and tried to wave us in. He didn’t realize he was waving at a robot taxi. I assume the Waymo algorithm doesn’t comprehend or obey friendly waves from other drivers.

So after sitting at that intersection for several minutes, unable to proceed, our taxi went into “emergency” mode. I assume that means it was “stuck” or “confused.” But shortly after that the car did what any smart human would do: give up and “make the block.” We turned right to avoid the whole situation and looped around to an easier intersection. Problem solved.

Riding with the Bot

I was giddy with the newness of it all once we got going. I looked over all the controls and nav screen as the taxi navigated down the road. The music was fine and the ac did well enough in the hot Phoenix sun. I changed the music to a different channel. I took a video with my phone.

The nav screen was interesting to watch. Similar to Teslas, the nav system displays much richer, interactive detail than typical car mapping systems. For what seemed like a couple hundred feet radius the screen displayed all the other moving vehicles and stationary objects, like buildings and trees, around us. According to Waymo, the Jaguar I-PACE SUV uses LiDAR, RADAR and cameras for navigation. That’s where this rich visual experience is coming from.

Stopping Power

Our Waymo ride went well enough except for one thing that stood out: stopping. My wife and I both agreed on this.

On several occasions, when approaching red lights or stop signs, the car did a “stutter stop.” Instead of gradually slowing like a human driver, the car would lightly jerk (stutter?) multiple times until it stopped. I’m usually immune to motion sickness, but even I got a little queazy at these moments. It happened several times throughout the stop and go drive.

I searched around and read a bit on ChatGPT about this stutter-stop behavior. It seems others have experienced it too. I’m sure this will get worked out in future tweaks to the system. I could not find much detail on the issue, but some Reddit users report a similar problem. ChatGPT thinks the problem involves the iterative nature of the car’s nav brain making multiple observations of its situation and the safety of the surroundings and so essentially it is “checking” for safety multiple times before fully stopping.

Onward

Waymo: We’re now providing more than 250,000 fully autonomous paid rides each and every week. The robotaxi future is here, and it’s powered by our generalizable Waymo Driver.

I love the ride-sharing experience, having tried Uber first among most (all?) of my friends about eleven years ago. I’ve enjoyed many conversations with taxi and ride-share drivers over the years. I would miss that if I went 100% driverless. Don’t get me wrong, I’m pro-technology. But I might not be completely ready to let go of the personal connection of the taxi driver.

Short phone video, with nav screen at the bottom.

About ten years ago (maybe more?) I would scroll through the Capital Factory event page and head down for some great times. And I still do! I have met many interesting and enjoyable people over the years. Their personal and work stories are incredible.

I’ve gotten jobs and met lifelong friends at events in Austin. I’ve especially enjoyed the Open Coffee events hosted by some fantastic Austin “super connectors” like Marc Nathan and Damon Clinkscales. Another recent favorite does 5 mile walks around Lady Bird Lake.

The AI Event

Later this month I’m hosting my own meetup. Topic? What else: AI!

All the cool kids are talking about it. Austin already has some great AI meetups, but I figured I would add one in North Austin to divert some of the cars driving in circles downtown searching for parking.

The meetup is for anyone interested in AI, whether you are an expert practitioner or want to learn more about it. Or your boss has started shouting, “Hey bot, send me that report, stat!” and you’re worried about job security.

In the meetup I want to discuss a wide range of AI topics, from practical tools used today to existential concerns for the (near?) future.

The meetup will be the fourth Thursday of every month at WeWork Quarry Oaks in North Austin - 5:30 PM to 7:00 PM. (Quarry Oaks, NOT the Domain. We like the Domain, but we’re not there).

Please join us - and take at least one car off the parking madness in downtown Austin.

After completing a 7-month deep dive course in AI & Machine Learning, I decided to try this contest. While it’s nowhere near the higher paying contests ($2.1+ million!), this one does have a $5,000 first place prize.

The Setup

After logging in and agreeing to competition terms, competitors have access to 1,500 real driving videos. The videos come from forward facing, dash-mounted cameras in cars and trucks on the road. Half of the 40-second videos show real accidents and the other half show normal driving. The trick is to train an AI algorithm to “learn” what elements of the video are predictors of a crash.

How Machines “Learn”

Computers don’t see videos like we humans do. Instead, the best a computer can do is to track the movement of all the tiny pixels that make up videos. In fact, those videos you watch on Netflix or any computer screen are actually a set of many still images shown at high speed, typically 30 frames per second, to simulate movement.

So our 40-second crash video is actually a series of 1,200 frames played fast enough that you see smooth motion instead of individual frames.

For our contest, the task involves tracking and measuring movements of all those pixels across hundreds or thousands of frames and building an association of those movements with “crash” or “not a crash.”

Features are Important

I’ll take a little creative license here for illustration. Imaging you are teaching someone to recognize different car models. All cars have wheels, headlights, tires, windshields and more. Those features are common across all cars, but vary wildly by brand and model of car. You could say, “Oh see that wheel, that’s definitely a Ford.” Or, “That tail fin is a key mark of a BMW.”

Essentially an AI model creates a feature set describing what it sees across all those video pixels. Now this get a little weird: the features that the model detects are abstract representations, not easily namable things we humans know like “headlights” or “bumpers.” The AI is detecting patterns which may or may not mean anything coherent to us if we inspected what it sees. To represent what it “sees” the AI creates long lists of numbers (aka vectors) that represent discreet features in each video. So, imagine this list going on for 2,048 elements:

[0.346
0.788
0.821
0.249
…
0.677]

That list of 2,048 numbers would represent a single video to the AI.

Now that we have a feature set (numbers) for each video, the AI begins to draw correlations. By doing some mathematical operations, it observes those features present in a crash versus those present in a normal driving. During this phase (known as training), the videos contain labels that clearly show whether a video has an accident or not. It’s sort of like flashcards for your friend who is learning car models: “Ford,” “Chevy,” “Toyota.” In this case, “crash” and “no crash.”

Survey Says

Once an AI has been trained we call it a “model” and use it for making predictions. In the contest, we actually do two types of predictions. In the first type, called validation, we know the right answer (crash or no-crash) and so at this phase we use data to reinforce the training, making it better and better over a series of iterations. So we show the AI a set of videos and ask it to make predictions. When it is right, great! When it is wrong, the AI uses that to incrementally adjust to make itself more and more accurate. Over and over and over, it trains.

In the second type of prediction, called “testing,” we show the AI entirely new data with the “crash” or “no-crash” labels hidden. In our case, we show it pre-crash videos and ask it to predict: crash or no-crash.

And that’s exactly how our contest works. We do lots of tweaking, training and testing and then finally we present our best estimates to the competition judge. The judges (in this case a computer) know the answers but we don’t so it’s a good test of how accurate our models are at predicting crashes.

The top scorer in the contest currently has an accuracy rate of over 90%. I was hovering around 20th from the top last time I checked. I’ve learned a ton about AI but I’m not in the upper echelons (yet!).

Saving Lives

Well that’s the big question, isn’t it? Will this technology reduce crashes and fatalities. My guess is that it will. Automotive safety has been on a long, steady improvement trend since the passenger car was introduced, as this data attests.

Sources

Moura, Daniel C., Shizhan Zhu, and Orly Zvitia. Nexar Dashcam Collision Prediction Dataset and Challenge. 2025. https://arxiv.org/abs/2503.03848.

After waiting in a line of hundreds (worth the wait) we walked through several event spaces.

Hands on

The bright, eager NASA people showed off several hands on exhibits. On the first floor we saw things like mini-mars rovers and plant health monitors.

Gadgets! The STELLA (Science and Technology Education for Land/Life Assessment) project uses spectrometers to detect plant health by inspecting the light reflected by plant leaves. These small, inexpensive gadgets are targeted at novices to help them learn principles of science by assembling their own hand-held spectrometers.

We also got to watch the obligatory cute kids playing with mars rovers.

NASA Talks

In the auditorium we caught the tail end of one talk and the full presentation on space telescopes. Dr. Amber Straughn is an Astrophysicist at NASA and spent fifteen years working on the James Webb Space Telescope (JWST). Straughn described what went into the planning, building and eventual launching of JWST on Christmas Day 2021. In fact, she was on a NASA call while enjoying Christmas at her mom’s house in Arkansas the day of the launch!

My favorite part of the whole day was hearing the tiny voices in the crowd during the audience Q&A. Several little future astronauts asked great questions, like “What is the Moon made of?” and “When will the Artemis go to the moon?”

Straughn also talked through several details about the multiphase plan to get people back on the moon. The first step has already happened, an uncrewed test flight of the ship that will eventually get people back to the lunar surface.

See the full list here: The List

Some featured events that really caught my eye below.

Museum of the Future House

Fri through Tues: March 7-11, 10:00 AM - 7:00 PM Congress Ave. & W 3rd St. An immersive experience exploring future possibilities, including themes like space travel, climate change, health, and how technologies like synthetic biology and AI can address global challenges. Capacity is limited to 250 people at a time. No RSVP required; SXSW badges get priority, followed by general admission. Statesman

NASA House: CreateSpace

Saturday, March 8, 10:00 AM - 4:00 PMAustin Central Library. An immersive experience where visitors don't just learn about space—they help shape it. Free and open to the public. SXSW

See the full list here: The List

The energy was great! I met lots of interesting people thinking about and building AI solutions.

Some big names like Meta, Red Hat and Lambda Labs were in the house.

I’ll definitely be back next month.

This one by Andrej Karpathy (former head of self-driving at Tesla) is fascinating and well-done. I’m about one hour in.

Check it out here:

February 22, 2025

AI Events in Austin: Your Guide to Regular Meetups and Communities

Austin AI Alliance Monthly Meeting
Capital Factory, Voltron Room
7:00 – 8:00 pm CT
*4th Wednesday of the month at Capital Factory in the Voltron Room, but may be changed due to holidays or room availability.

Austin Deep Learning
Capital Factory, Voltron Room
7:00 PM to 9:00 PM CST
*1st Tuesday of the month  

AITX Monthly Meetup
Capital Factory on the 5th floor.
5:30 – 7:30 pm
*4th Tuesday monthly

The post AI #5 – AI Meetups in Austin, TX appeared first on Steve Ward Media.

• December 18, 2024

Truly Usable AI for Documents

Google is behind in the AI race (hard to believe, I know), with OpenAI’s ChatGPT leading the pack.

But, have you checked out NotebookLM? The capabilities are mind blowing. You should give it a try. Find a document or PDF which made you think “ugh, I don’t have time to read this thing.”  Hop over to NotebookLM and upload the doc. instead.  

Even Cuts Through Insurance Legalese

A perfect use-case for NotebookLM came up for me recently. After starting a new job a couple months ago, lots of mail started showing up. One letter was a dense 5-pager full of insurance legalese. Do I need to do anything?  Is the insurer telling me there is a problem?  Who knows? And who really wants to sit down and read every lengthy form letter that comes in the mail?

So I simply scanned that letter and uploaded it to NotebookLM.  The first thing I got was a quick overview summary of the document. NotebookLM can do this whether it’s five or 95 pages. See the nearby image for my insurance example. Next, I was able to “ask” the document questions.  So I asked if I needed to take any actions.  I also asked if there was a problem I needed to deal with.  Both came back with clear explanations that I did not need to take any action and reasons why. 

And a “Podcast” Too?

If that wasn’t enough, NotebookLM will also create a podcast for you. Yep. A podcast. 

After ingesting file attachments, NotebookLM has an optional feature which creates a podcast-style discussion. Two people “discuss” the content of your document, providing thoughtful summaries and highlighting special areas of focus. I’ve tested it many times and found it especially useful to review lengthy content while driving. 

The voices are realistic-sounding, with one male and one female. They casually discuss your content like intelligent, informed people. They even include human ticks, like “umm” and raising and lowering of volume.  

Amazing.

Give it a try at notebooklm.google.com and let me know what you think. 

The post AI #4 – Google’s Must-See NotebookLM appeared first on Steve Ward Media.

• December 13, 2024

Have You Seen It?

That’s exactly what my good friend, Sam, said as he walked into the office today.  “Have you seen it?”

See it first and then I’ll add some comments.

VIDEO: My Friend Working With ChatGPT

Just whoa! Are you getting this? ChatGPT with Advanced Voice and Vision can see, hear and follow instructions based on what it “sees.”

The uses seem endless.  The other use Sam showed me was of ChatGPT helping him practice his acting part in a stage play. Pointing his phone at the printed page of dialogue, ChatGPT was able to read back and forth with Sam interactively!

The post AI #3 – How’s That Possible? appeared first on Steve Ward Media.

• December 10, 2024

The new game in town: is it real or AI? 

This game has been going around for a couple years now.  But now that AI has gotten so much better the question is much more interesting.  

A few weeks ago Scott over at Astral Codex Ten had a post on this theme.  He posted a quiz asking participants to guess which of 50 artistic images were human or AI-generated.  The results are fascinating.  The official quiz is over, but you should try it out anyway.   

The Latest in AI Video

That brings us to new from Sora, the text to video engine by OpenAI (makers of ChatGPT). You have to check it out. I’ve only looked around for a few minutes, but man – it is amazing.

I love dogs, so this one is A+ for me.  This seahorse is amazing too. 

And of course there is a video from OpenAI to go with the launch.  See it here.  

The post AI #2 – Is it Real or AI? appeared first on Steve Ward Media.

• December 6, 2024

AI is ubiquitous in the news. So much is happening that even experts have a hard time keeping up.  Here I’m sharing the sources I follow for the latest news, interesting debates and technology updates.  

Podcasts

Last Week In AI – founded by a Stanford grad with a PhD in AI.  Hosted by super smart duo, Andrey Kurenkov and Jeremie Harris, who share headlines and deeper stories.  

The AI Daily Brief – quick daily hits. Very interesting and informative. 

The Super Data Science Podcast – more focused on the field of data science but also has some great interviews, for instance recently with Andrew Ng of Deepmind.AI and Coursera. Fantastic host Jon Krohn.

Newsletters

Don’t Worry About the Vase by the incredible Ziv Moshowitz, a fascinating and prolific writher. Economist and Magic: The Gathering gaming pro (as in played as a professional).

Top AI Chat Models (LLMs)

ChatGPT – often considered the top chat LLM. Part of OpenAI offerings, with large investments from Microsoft.

Claude – Advanced LLM by Anthropic, a company started by OpenAI alum and focused on safety/alignment. 

Perplexity – A powerful model which places reference links for all of its output, emphasizing sourcing for its LLM responses. 

XAI – Elon Musk’s AI from X.  

Gemini – Google’s frontier model.  

NotebookLM – a unique take on LLMs. You can feed in up to 50 large documents and even YouTube videos and ask questions about the data contained therein. 

–

Courses

DeepLearning.AI (Coursera) – deeplearning.ai was founded by Andrew Ng, one of the pioneers in AI and machine learning education. Many of the courses, like the highly-regarded Deep Learning Specialization, are hosted on Coursera. 

University of Texas – McCombs School of Business – Post-Grad Program in AI and Machine Learning – the seven month course I am taking through UT. I’m about five months in.  Highly recommend. I’m thoroughly enjoying the learning journey in this course. 

The post AI #1: AI Resources appeared first on Steve Ward Media.

• September 12, 2024

Explosive Projects!

Sitting across the booth from my co-worker, we eased into a pleasant lunch break conversation. We were in town for the construction of a large new corporate campus. The site would be home to thousands of employees doing lots of different jobs. As usual, we were on a tight timeline.   

Then my mobile phone (probably a Nokia candy bar style!) buzzed.  “Steve, uh, you have to come see this,” said the person at the other end.

Big projects come with a myriad of issues, but this urgent call had my mind racing.  ‘What could be the hangup this time?’ I wondered. 

We quickly finished up lunch and headed to our cars. 

I didn’t know what to expect when I arrived.  As I pulled up to the busy construction site, the usual big, yellow front end loaders and backhoes rumbled around. It seemed pretty normal so far.  

Then I stepped out of my rental car.

Standing alongside my co-workers and the general contractor’s team, we stared down into what can only be called a giant crater. Below us a 15-foot diameter hole in the ground stared back. Rock and rubble were strewn everywhere.  

You see, the geology of the area made dynamite a useful method of clearing and site prep. It turns out that dynamite had blasted away our recently installed, highly delicate fiber optic cable for the campus. From one side of the hole I saw a ragged, ripped end of a black-clad fiber cable. Several feet across on the other side was its match, similarly ripped and ragged.

That fiber cable was definitely on the critical path, as we project managers say.  Without it, the entire schedule and opening deadline was in jeopardy. That fiber was designed to handle data and communications for the entire multi-acre site, carrying thousands of calls and millions of bytes of data daily.

Thankfully, our experienced team had already put in motion the fix we would need. Calls were made to the telco, discussions we had, solutions were being devised.  

In the end we were able to react and bring in the telco and construction teams to repair the fiber.  It was a nerve-racking part of an aggressive project.    

Steve Ward

Steve Ward is a father, husband, sailor, writer and project management expert. He lives in the hills of west Austin with his wife, daughter and two big golden retrievers. His labor of love is sailATX, a sailing excursion co. on Lake Travis.

The post Explosive Projects! appeared first on Steve Ward Media.