Research - Fundamental

Building the next generation of AI for enterprise prediction

Fundamental is a frontier AI research lab building foundation models for the world's structured data

Our research tackles one of deep learning’s most complex challenges: building general-purpose models that can learn across the real-world tabular data that powers businesses and industries

Rethinking Foundation Models

Foundation models transformed how machines understand language. We believe the same fundamental shift is coming to structured data.

Our research explores how models can learn across diverse datasets, generalize to unseen prediction problems, and operate at real-world scale

Go Deeper

Our Research

We’re building the foundations for a new class of AI

Our research spans the fundamental learning problems, architectures and large-scale systems required to build foundation models that can learn from the complexity of real-world structured data

01

Meta Learning

02

Architecture Research

03

Generative Modeling

04

Large-Scale Systems

05

Code Generation

06

Structured Data

Meta Learning

Building models that learn how to learn across datasets, domains and prediction problems

Real-world tables rarely share the same schema, semantics or statistical structure. We study how knowledge learned across diverse datasets can become useful prior knowledge for entirely new problems.

Learning transferable representations across heterogeneous datasets

Rapid adaptation to new datasets, schemas and prediction tasks

Understanding what knowledge can generalize across domains

Meta Learning

01

Meta Learning

02

Architecture Research

03

Generative Modeling

04

Large-Scale Systems

05

Code Generation

06

Structured Data

Meta Learning

Building models that learn how to learn across datasets, domains and prediction problems

Real-world tables rarely share the same schema, semantics or statistical structure. We study how knowledge learned across diverse datasets can become useful prior knowledge for entirely new problems.

Learning transferable representations across heterogeneous datasets

Rapid adaptation to new datasets, schemas and prediction tasks

Understanding what knowledge can generalize across domains

Meta Learning

Talent from the frontier of AI research

Researchers, engineers and operators from some of the world's leading institutions and technology companies - brought together to build a new category of AI

Berkley

Bridgewater

DeepMind

MKIT

Oxford

Palantir

Standford

Berkley

Bridgewater

DeepMind

MKIT

Oxford

Palantir

Standford

Join Us

Publications

Developing Foundation Models for Real-World Tabular Data

Marta Garnelo & Wojciech Marian Czarnecki

A research manifesto for building general-purpose foundation models that learn across tabular datasets and adapt to new prediction problems

Read the White Paper

Developing Foundation Models for Real-World Tabular Data

Marta Garnelo & Wojciech Marian Czarnecki

A research manifesto for building general-purpose foundation models that learn across tabular datasets and adapt to new prediction problems

Read the White Paper

Developing Foundation Models for Real-World Tabular Data

Marta Garnelo & Wojciech Marian Czarnecki

A research manifesto for building general-purpose foundation models that learn across tabular datasets and adapt to new prediction problems

Read the White Paper

Why Large Language Models Fail at Tabular Prediction

Marta Garnelo & Wojciech Marian Czarnecki

An examination of why architectures designed for language struggle with tabular prediction - and what fundamentally different capabilities these problems demand

Read the White Paper

Why Large Language Models Fail at Tabular Prediction

Marta Garnelo & Wojciech Marian Czarnecki

An examination of why architectures designed for language struggle with tabular prediction - and what fundamentally different capabilities these problems demand

Read the White Paper

Why Large Language Models Fail at Tabular Prediction

Marta Garnelo & Wojciech Marian Czarnecki

An examination of why architectures designed for language struggle with tabular prediction - and what fundamentally different capabilities these problems demand

Read the White Paper

What LLMs learn (and don’t) from tables

Marta Garnelo & Wojciech Marian Czarnecki

An ICML presentation examining why a frontier LLM's tabular predictions rely on memorization rather than genuine learning, and collapse as the number of columns grows

Read the Presentation

What LLMs learn (and don’t) from tables

Marta Garnelo & Wojciech Marian Czarnecki

An ICML presentation examining why a frontier LLM's tabular predictions rely on memorization rather than genuine learning, and collapse as the number of columns grows

Read the Presentation

What LLMs learn (and don’t) from tables

Marta Garnelo & Wojciech Marian Czarnecki

An ICML presentation examining why a frontier LLM's tabular predictions rely on memorization rather than genuine learning, and collapse as the number of columns grows

Read the Presentation

First Principles

Conversations about the ideas shaping the future of AI

First Principles brings together Fundamental's research team and leading thinkers in machine learning to explore the questions and ideas shaping the field

Join Us

We're looking for exceptional researchers and engineers who want to work on foundational problems in machine learning - and turn those ideas into real-world systems

See Open Positions

Join Us

We're looking for exceptional researchers and engineers who want to work on foundational problems in machine learning - and turn those ideas into real-world systems

See Open Positions

Join Us

We're looking for exceptional researchers and engineers who want to work on foundational problems in machine learning - and turn those ideas into real-world systems

See Open Positions

Join Us

We're looking for exceptional researchers and engineers who want to work on foundational problems in machine learning - and turn those ideas into real-world systems

See Open Positions

Terms & Conditions

Large Tabular Models

Terms & Conditions

Large Tabular Models

Terms & Conditions

Large Tabular Models

Large Tabular Models

Talk to Sales

Talk to Sales