Advising Statement for Prospective Students
My research interests and advising priorities are changing in response to a broader transformation in AI.
I believe AI research is entering a new stage. The most meaningful advances will increasingly come from real-world deployment, real user needs, real data, and closed-loop systems that continuously improve through actual use. In other words, I am less interested in research that only optimizes for existing datasets, public benchmarks, or isolated paper-style problem formulations. I am more interested in AI systems that enter real settings, encounter real constraints, generate real feedback, and build long-term data flywheels.
As a result, my current research is highly focused on real-world AI systems, industrialization, data loops, and durable technical capabilities. This path is different from the conventional academic model centered primarily on rapid publication. Many of the problems I care about do not come with existing papers to follow, mature templates to copy, or clearly defined benchmarks to optimize. In many cases, we need to define the problem ourselves from real-world observations.
This also means that my expectations for students are different.
If your main reason for applying to my group is one of the following, my group may not be the right fit:
- You mainly see the PhD as a stepping stone toward short-term industry compensation.
- You mainly want a PhD credential in order to pursue a conventional academic career.
- You mainly want to use the PhD years to accumulate internships and resume signals.
- You mainly want to publish many papers quickly and treat publication count as the central goal of doctoral training.
These goals are understandable, and there is nothing inherently wrong with them. They are simply not the direction I am currently most interested in supporting.
I am looking for students who are willing to work with me on real-world problems. This requires comfort with uncertainty, patience with long feedback cycles, and the ability to make progress when there is no mature template to follow. You should care about whether a problem is truly important, whether the work creates durable value, and whether it builds capabilities that are difficult to replace.
In particular, I value the following qualities.
Long-Term Orientation
You should care about long-term accumulation. The key question is not whether a project can be quickly packaged into a paper, but whether it addresses a real problem, creates durable value, and builds a meaningful technical moat over time.
Focus and Resilience
Research grounded in real settings is often uncertain, slow, and messy. It may not immediately produce publishable results. I expect students to remain calm, focused, and persistent, and to work on what is right rather than constantly shifting toward what seems easiest to reward in the short term.
Vision and Conviction
You should have, or be willing to develop, your own view of where AI is heading. You need to understand why real data, real systems, and closed-loop deployment matter, and you should not be easily pulled away by short-term competition or conventional status games.
Ability to Execute
I value students who can turn ideas into working systems, datasets, experiments, and deployed workflows. Real-world problems do not automatically arrive as clean paper topics. They need to be observed, defined, built, tested, and refined.
This is not the easiest path, nor is it necessarily the most efficient path by short-term academic metrics. But I believe it is closer to where the most important AI research is going.
If this direction resonates with you, and if you are willing to commit to real-world AI problems over a long horizon, you are welcome to reach out.