Meet Jonathan Frankle, who taught programming to law students, then a 'lottery ticket' idea led to a $1.3 billion exit and Databricks' top AI job

Not everyone who builds a career in artificial intelligence takes an easy and simple path. But, for Jonathan Frankle things were different. Frankle's journey in AI took an unusual path. He later went on to become the Chief AI Scientist at Databricks. Before that, he worked on computer science research and taught programming to law students. He later developed a research idea. This he developed during his PhD that later became known as the Lottery Ticket Hypothesis. It helped establish his reputation in machine learning.

He earned his PhD from the Massachusetts Institute of Technology (MIT). There he studied how neural networks learn. He also worked with researcher Michael Carbin on a question that became important in AI development.

What is the Lottery Ticket Hypothesis?

Frankle and Carbin published a paper in 2018. It introduced the Lottery Ticket Hypothesis. The idea was simple. It was that a large neural network could contain smaller networks capable of learning just as effectively when trained on their own.


They called these smaller networks as "winning tickets." It suggests that certain connections and starting values within a neural network could be particularly useful for learning.

This offered a new way of thinking. The paper later received the Best Paper award at the International Conference on Learning Representations in 2019.

ALSO READ: Meet Ali Partovi, man who bet on young talent and backed Cursor and Kalshi; his $150 million Neo fund reached $1.2 billion in value by the end of 2025

Frankle's journey from research to business

Frankle later became part of the founding team at MosaicML. It was an AI startup established in 2021. There, he served as its chief scientist. He also helped lead research into making AI model training more efficient.

MosaicML had one focus. It was on helping businesses build and train their own machine learning models rather than depending entirely on ready-made AI services. The technology was designed to make training large models more accessible and cheap.

ALSO READ: Meet Samantha and Kevin Dwoskin, couple who started a mac-and-cheese side hustle in 2023; about 20 months later, Boss Cow had crossed $1 million in annual revenue

Why did Databricks acquire MosaicML?

Databricks announced a big in June 2023. It revealed an agreement to acquire MosaicML in a transaction valued at around $1.3 billion. It included retention packages. The deal was completed in July 2023

Databricks wanted to combine its existing approach with MosaicML's technology.

Notably, Frankle continued his work there and became Chief AI Scientist and led research focused on training neural networks more efficiently.