A new framework uses synthetic red-green-blue (RGB) imagery, a small set of real annotations, and generative adversarial network (GAN)-based cross-modality alignment to improve semantic segmentation of crops and weeds in thermal images. 

When Miriam Friedel started working at Capital One in 2020, generative AI was still largely a research curiosity. By the time she left in 2026 as vice president of machine learning engineering, AI tools were writing code, building applications and reshaping how people work. Friedel has since joined the University of Virginia School of Data Science as an associate professor of practice.

In an interview, Friedel says learning data science is still relevant because AI can help explore data, but it can’t tell you whether you are solving the right problem in the first place. She also warns that AI tools are often sycophantic, telling people what they want to hear, which is a problem when making important decisions.

For students, her advice is to use AI for things they can verify the answer to. “Use AI like a TA, like a sparring partner, but you have to actually do the work,” she says. She adds that getting frustrated and pushing through hard problems is how people learn and innovate.

For job seekers, Friedel says employers are not looking for people who simply paste in AI output. “Typing a prompt into ChatGPT, waiting a few minutes, and cutting and pasting the answer without review is not a skill,” she says. Employers want people with the discernment to know when AI speeds up the work and when it is getting things wrong, and who will take responsibility for the risk when AI-written code goes into use.

Looking back on her time as an executive, Friedel says the hardest problems were not technological. They were figuring out the right problem to solve, executing on it and working with other people.