The Left Brain of AI: Why Large Tabular Models Signal A Renaissance For Data Scientists

The Job No One Describes in the Job Description

What I didn't expect was how human the job would turn out to be.

At DataRobot, I was extremely privileged to spend years sitting across from nurses, traders, underwriters, call-center operators - the frontline workers who ensure the critical operations of their organizations keep running. They carved time out of genuinely packed days to teach me their daily lives, and I understood then that the job was way more than running a `.fit` and `.predict`. To make real impact, you need to deeply understand your users, their world, and what decisions they're actually trying to make and then bring them exactly the right information, in advance, to make those decisions easier.

That's what data science is, to me, at its best.

The Ground Shifts

Then November 2022 arrived, and it shook me. After a decade spent deep in statistical and machine learning, watching the gen AI explosion unfold felt like the ground shifting beneath my feet. Was the work I'd spent years building suddenly irrelevant? Were structured data and predictive modeling about to become yesterday's news?

I joined Mistral AI to figure out how to make GenAI reliable enough for the real world, and essentially went back to school: pre-training, post-training, tool integration, workflow architecture. A fast, humbling education in a technology that was pulling all the attention, and rewriting the rules.

But the more I understood it, the more the tension became impossible to ignore.

The Wall

LLMs are genuinely remarkable at handling unstructured thought - language, nuance, creativity, imagination. But strip it back, and they remain powerful next-token predictors, and that's precisely where they hit a wall. What they're not built for is the other half of the enterprise: the approvals, the risk assessments, the fraud flags, the supply chain calls. Those decisions are binary. Deterministic. They live in rows and columns, not paragraphs. And they need to be not just correct, but auditable and defensible.

The Left Brain Awakens

What excites me is that tabular data is now having its own GPT moment. A new generation of models - borrowing transformer-based architecture from the LLM world are being purpose-built for structured prediction. If you've spent years tuning XGBoost pipelines, pay attention — this isn't another boosting variant. It's a genuine paradigm shift: pretrained tabular models that transfer-learn across datasets out of the box. 

My friend Dean Wang put it perfectly: LLMs are the right brain of AI, creative and associative. Large Tabular Models are the left brain - logical, structured, built for decisions that actually have to hold up under scrutiny. You need both. Most enterprises are only running on one.

A Head Start, Not a Setback

For the data science community, this is very good news. If you spent years in the trenches of predictive modeling, you are right to be excited - those skills are back at the center of something important. You're not behind. You have a head start.

Fifteen years ago, I walked away from finance because I believed you could build better theories from data than you could impose on them. That conviction hasn't changed. What's changed is that the tools are finally catching up to the ambition.

The future of AI needs both brains working together. And if you're a data scientist who's been quietly wondering whether your skills still matter - they do, and the next few years are going to prove it.

For those who want to go deeper on the technical case, our research team's whitepaper is here: Whitepaper.PDF

Alexandre Gerbeaux
Head of Applied AI at Fundamental