Lightfast is an applied AI research lab.
We study a central question across science and engineering: how should people, machines, and artificial intelligence work together to make discoveries? Research comes first. Alongside data and environments, we develop new methods of our own, and new ways to study and repeat discovery.
People, Machines, and Artificial Intelligence
People bring purpose and domain knowledge, and retain responsibility for decisions and their consequences. Machines sense and act in the world; artificial intelligence can interpret context and connect human intent with what machines do.
People also guide AI while it works, in an experiment as much as in a piece of art. We study how to join these roles in systems that keep people directing the work.
Studying Breakthroughs
Breakthroughs in science are a product of creativity. A breakthrough is a new way of framing or solving a problem, and the moments that define it are rarely studied directly; they are our core interest.
One way we study them is to put many agents on one hard problem, where a few may find an answer and most will not. We study the whole set of runs, the failures included, to understand what happened in the ones that succeeded, and to turn the study of breakthroughs into a science.
Data and Environments
Models fall short in understanding human taste: what is worth trying, and what is good. Taste belongs to science as well as to art, and for a scientist it is how they think and work during an experiment.
Part of the shortfall comes from the shape and curation of the data models learn from, and that data is our focus. We curate the data we think does most to build that understanding, and build reinforcement learning environments, digital and physical, with the checks that go with them. We are the curation layer above what goes into training.
Research Alongside Teams
We work alongside teams across research and industry to discover the use cases that push the frontier. Together we find where models fall short in a specific kind of work, design the data or environment that could help, and test whether models improve. Teams set their own objectives and define success.
What generalizes becomes research, datasets, environments, tools, shared infrastructure and, in time, products.
Work With Lightfast
If your work could change what people and machines discover together, we would be glad to hear what you are working on.
