NVIDIA sells $200 billion worth of chips every year. But Jensen isn’t stopping there. NVIDIA recently dropped its most powerful AI model ever. 

This morning I tested Nemotron 3 Super, NVIDIA’s most sophisticated model. Nemotron is designed to let us build and run powerful agents cheaply.

Nemotron does well if you give it data (also known as Retrieval Augmented Generation, or RAG). But it struggles with complex jobs and finding data on its own. 

Let me show you where Nemotron can help and where you’re better off looking elsewhere…

Round #1: Finding the Next Unicorn on Product Hunt

Several times a week, I troll Product Hunt looking for the next great startup. Can Nemotron automate this? 

Nemotron wouldn’t help, saying that it can’t browse the internet. For a model designed to create agents, this seems like a huge problem! 

You can add a web browsing tool yourself. But it would be much easier if Nemotron included it automatically. 

What’s more, Nemotron’s knowledge cutoff is July 2024, nearly two years ago. This makes it harder to build useful tools. 

I love Jensen, but I have to give this round an F. 

Round #2: Researching Drone Technology

Earlier today, I chatted with a founder making very innovative drone motors. His motors don’t use any rare earth minerals.

English major that I am, I don’t understand how existing drone motors use rare earths. Can Nemotron explain?

Nemotron gave a great breakdown: the rare earths are used in a magnet that the motor needs to function.

Nemotron’s explanation was a little technical. I asked a few follow-up questions, and it did a good job of breaking things down more simply.

But there’s one big problem: Nemotron didn’t cite any sources. The information sounds plausible, but I can’t verify it.

This makes it hard to rely on Nemotron. I’m giving this round a B-minus. 

Round #3: Automating Intros to Startups

Yesterday, I met a really cool biomanufacturing startup. Which investors in my network should I introduce them to?

I fed Nemotron my spreadsheet of 75 top investors. Nemotron did a great job of looking through the list and finding investors interested in deep tech.

When I give Nemotron all the data it needs, it parses it very well. You can take advantage of that in applications you build, RAG-ing Nemotron and getting outputs cheaply.

I’m giving this round an A. 

Wrap-Up

I’m giving Nemotron a C- overall. 

Nemotron really deserves two grades: an F for tasks requiring web search and an A for parsing data you provide. 

Nemotron is way behind frontier models. But because it’s cheap to run, Nemotron could be a good choice for your application. 

Just be sure to give it very specific instructions and lots of custom data. Then, you can get the output you need at lower cost. 

I’m excited to see what Jensen does with Nemotron. Tenacious as he is, he just might score a come-from-behind victory.

More on tech:

Google Drops New Gemini macOS App — A for Easy Tasks, F for Complex Jobs

Grok Imagine Quality Mode: Jaw-Dropping Images, But Text Still Struggles

Meta’s New Model is Worth $60 Billion — But Struggles in Real World Testing

Save Money on Stuff I Use:

Fundrise

This platform lets me diversify my real estate investments so I’m not too exposed to any one market. I’ve invested since 2018 with great returns.

More on Fundrise in this post.

If you decide to invest in Fundrise, you can use this link to get $100 in free bonus shares!

Wispr Flow

I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.

My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.

Get a free month of Wispr Flow Pro here!