Across the episode, Ben explains how Two Sigma researchers think about transforming raw data into predictive features, why creativity and scientific rigor are both essential to the research process, and how AI can help researchers explore more ideas while keeping human judgment at the center.
Read on for key moments from the conversation, and listen to the full episode below.
Features are facts about the world worth noting
Ben begins by explaining what Two Sigma means by a “feature” in quantitative research: facts about the world that are worth studying and economically meaningful. That could be something familiar, like how a stock price moved over the past week, or something more specific, like how much news coverage a company is receiving, whether an analyst upgraded the stock, or when a press release was issued.
The key is not simply collecting data, but using human knowledge and intuition to identify which details may be economically meaningful enough to test.
Turning raw data into differentiated ideas
The source of edge in quantitative investing has shifted as more data has become widely available. While proprietary data still matters, Ben explains the ability to transform data into useful, creative and differentiated features has become increasingly important. Researchers might begin with a common data set, but the real value comes from asking unusual questions and finding patterns others may not have considered. To Ben, this human ingenuity layer is central to how Two Sigma thinks about feature research.
A good part of what’s creating this alpha is certainly the human ingenuity layer of how to convert piles of bits into features that are interesting, meaningful, clever, and maybe that you think others haven’t thought of.”
Building on a shared research platform
Two Sigma’s research culture is one built around shared contribution rather than isolated teams competing against one another. Researchers use shared feature libraries and collaborate to build models together, leveraging the data moat and compute power of the shared platform.
Ben notes this structure encourages teams to optimize for the firm’s overall predictive capability, rather than only for a local portfolio or individual group. The result is a research environment where ideas can compound as they move across teams.
Searching for orthogonal and novel signals
A recurring theme in the conversation is the importance of orthogonality. Ben explains that it is relatively easy to create features that are correlated with signals already in use, but those are not usually the most valuable. The goal is to find new predictions that are meaningfully different from what already exists. That search for novelty helps preserve diversity across models and strengthens the overall research process.
Our job is at its core every day to come up with orthogonal and novel predictions. It’s kind of easy to come up with things that are correlated to what we’ve done in the past. And that’s the lowest hanging fruit, right? And so almost by definition, we wake up every day as an alpha modeler searching for orthogonality.”
Combining creativity with the scientific method
Feature research is both creative and scientific. Researchers need to generate novel hypotheses, but they also need to test those hypotheses rigorously and avoid rationalizing results after the fact. If a feature does not behave as expected, the discipline is to recognize that and move on, rather than force a story around the data. At Two Sigma, this combination of imagination and rigor is what allows creative ideas to become useful research inputs.
Using AI to expand what is researchable
Drawing on his background in natural language processing, Ben explains how large language models are expanding what can be studied as text. Historically, NLP focused on turning language into numbers; now, LLMs can turn numbers and other structured information into language that can be analyzed. For a firm that has spent more than 15 years studying text as a source of market insight, that shift has opened up a much broader universe of researchable data.
Amplifying human originality rather than replacing it
Ben emphasizes that AI should amplify what makes researchers distinctive, rather than push them toward the same answers. In quantitative investing, where originality and orthogonality are essential, the value of AI lies in helping people scale their own ideas, judgment and ways of thinking. That also means the design of AI tools matters: if automation becomes too generic, it could reduce the diversity of thought that strengthens the research process.
For Ben, the opportunity is to build tools that make researchers more creative and effective, not more interchangeable.
We see AI as an amplification tool of our originality.”