Noah on AI: Issue 20

Let’s get started—plenty to cover this week.

Gemini 3 is Here and Google wins the week! The model has been very highly regarded across the industry, and received notably high praise by Salesforce CEO Marc Benioff. Here are the positives:

Gemini 3 was integrated into Google Search on day one, and with the release, Alphabet stock (Google’s parent co.) was catapulted to an all time high. While the benchmark numbers are impressive, it does seem like the model was specifically trained for benchmarking in some cases.

I’ve played around with it a bit and haven’t seen a huge difference to the previous model on the LLM side. With that being said, there are some notable improvements to the API, the image model (Nano Banana Pro), and the advanced thinking model. Overall, an impressive launch and I remain very bullish on Google.

Who’s moving, who’s stalling, and who’s climbing: Ordered by production this week.

Interview Outline: This panel announces a major AI infrastructure partnership between the Kingdom, xAI, and Nvidia to fuel the new "Intelligence Age." Elon Musk and Jensen Huang discuss the transition to generative AI factories and humanoid robotics.

About the Interviewees: Hosted by Saudi Arabia's Minister of Communications and Information Technology, this panel features two of the world's most influential technology leaders. Joining him is Elon Musk, the CEO of Tesla and SpaceX and founder of xAI, alongside Jensen Huang, the founder and CEO of Nvidia, whose pioneering work in graphics processing units (GPUs) and accelerated computing has become the fundamental bedrock of the modern AI era.

Interesting Quote: “Work will be optional in 10-20 years. It will be like playing sports or a video game. If you want to work you can, if you don’t want to you won’t have to.”

My Thoughts: Interesting interview from two people atop the AI landscape. Elon seemed more optimistic while Jensen seemed to lean towards short-term knowns. I think people will always work in some capacity, but I can absolutely see the current 9-5 corporate structure being looked back on in 10-20 years as something from an outdated world. The concept of data centers in space makes too much sense - 100% see that in the cards. Watching the growth curves of these robotics companies and major AI companies even over the next 2-3 years should paint a much clearer picture about what the long-term future will look like. There are so many unknowns right now that it’s near impossible to accurately predict what the world will look like in the mid-long term future.

Difficulty: Mid-Level

We’ve talked about video tools and we’ve talked about images, now let’s talk about product videos. For anyone selling a product and wanting to use AI to demonstrate it, this is how it’s done.

Nano banana → Video model

Depending on the realism you’d like, there are tons of different video models you can use, and some we’ve covered in the past. In this example, we’ll use Gemini’s Veo3. To make a product video you need an image, a background, and an idea for a scene.

In our example we’ll use a Coca Cola can as the product, a couple different skiers on a mountain as the background, ElevenLabs audio for a voiceover, and iMovie to stitch the clips together.

I’ll also be using previous Practical use Cases: “Dual prompting” and “JSON prompts” to help craft the scene.

Here’s our starter images:

Then I asked nano banana to add the coke can into each skiers hand as seen here:

For both the images, and the video, I had ChatGPT write a prompt. One to input the coke can, and the other to instruct the skiers to take a sip while going down the mountain. I then pasted said prompt into Veo3 with the respective images.

Next, I picked an ElevenLabs voice to narrate the scene based on a short transcript. I took that audio file, added it to iMovie with the videos, added a Nano-Banana generated image and here was the result:

The cool thing about this isn’t the video. The one I put together is essentially AI slop with stock images and coke cans. The cool piece is the ability here to incorporate any background / characters alongside any product, and create a video from the two.

The best way to improve the ad above is spending a bit more time to refine some of the prompts, ensuring the starting images are clear, and testing a few different video models until a perfect medium is reached.

If you want to try it yourself, check out this tutorial ⬇️

The Problem: Most home robots struggle because they’re trained in sterile lab environments or rely on industrial methods. They don’t adapt well to messy, real‑world households.

The Solution: Sunday’s robot Memo uses a unique training system: humans wearing “Skill Capture Gloves” perform everyday tasks in real homes, generating rich data that Memo learns from. It can then autonomously handle chores like clearing tables, loading dishwashers, folding socks, and even making espresso in homes it’s never seen.

The Backstory: Founded in 2024 by Stanford‑trained roboticists Tony Zhao (CEO) and Cheng Chi (CTO), Sunday has built a team from Tesla, DeepMind and other high‑end AI/robotics labs. They’ve collected millions of real household chore episodes and built Memo with a rolling base (not legs) for safer, more stable motion. 

My Thoughts: I’m surprised more robotics companies are not using skill-capture gloves or similar methods to train on tasks requiring nimble movement. Very impressed at the speed in which Sunday Robotics is moving, and from their demos it seems like the technology is there. The robot itself also looks a lot better on the friendly-creepy scale than some of the others. Definitely a robotics company to follow.