ChatGPT Doesn't Just Do Chat: How to Build Real Data Models Inside ChatGPT o3—works for Claude and Gemini as well

ChatGPT o3 and other advanced models have become Everything Engines—using hundreds of tools to get complex work done—this post shows you how to build a real data model with an LLM!

Ok this one has been brewing for awhile. I’ve been feeling this huge looming gap between the way I use ChatGPT and Claude and Gemini and other LLMs and the way lots of my colleagues use these tools.

Fundamentally, with the launch of Claude 3.7, Gemini 2.5 Pro, and ChatGPT o3 these LLM chatbots are becoming Everything Engines. But here’s the catch: you can’t tell, because they don’t look different.

It is really hard to explain, so I decided to show it instead. I built three real models inside ChatGPT o3 that I road-tested on real data. It’s not just an LLM simulating a model. Because o3 is calling python and other tools and actually designing and building an actuarial model it can explain for these three challenges.

What are they? Well they’re the most interesting questions I could think of that I wanted to look at more closely from a data perspective.

1. Will this AI product launch succeed?

2. What are the odds I’ll buy a house in my zip code?

3. Is that layoff really “because of AI”?

We talk about all of these topics in the news all the time and mostly it’s a bunch of opinions. What if we built a model to represent what we know about the underlying data in a more rigorous way? Before, that would have taken a whole data team and lots of patience. Now, it’s just a smart chat away.

I hope you enjoy these three models, and I hope you get a sense of how easy it is to build more complex (not language) outputs in these advanced Everything Engines.